ADE & Agent-Orchestration Tracker

T3 Code now has LaTeX math rendering as an opt-in toggle in settings, per Theo replying to a user who asked for it. This reverses his Oct 12 "never native LaTeX" call by keeping it off by default so normal markdown isn't affected.

Theo - t3.gg@theo

View on X ↗

@Josikinz It’s a toggle in settings now (replying to: "@theo Please add latex rendering into the UI while ur at it Theo, a mathematician friend had to move because of that")

♥ 0⟲ 0💬 0👁 64

OpenCode shipped experimental background compaction (experimental.background_compaction, v2.1.4+): at 85% of the compaction threshold it writes the summary in the background from the cached prefix while the session keeps going, then swaps it in instantly at the threshold, keeping newer turns verbatim. OpenCode's demo claims the median compaction pause drops from 55 s to 0.002 s (p90 84 s to 11 s). Stale summaries are discarded and redone, sessions under ~64k tokens are skipped, and the desktop app shows one quiet Compacting/Compacted divider (PRs #57444, #58294, #58996; 121 likes, 5.5k views).

F

Filip@Neriousy

View on X ↗

background compaction just landed in @opencode as a experimental setting go try it out and tell us what you think

♥ 121⟲ 8💬 15👁 5,477

"Share with guest" (nightly, in T3 Connect): send a coworker a link to a thread; they can add context or messages, but the owner decides what goes to the agent, like a group food order.

T

Theo - t3.gg@theo

View on X ↗

New beta feature in T3 Code: "share with guest" I'm so tired of copy pasting from a thread, sending it to a coworker, waiting for them to respond, screenshotting the context and sending it back to my agent. Now I just send them a link. It's a silly place to start, but I was heavily inspired by the Uber Eats/Doordash group order flow. If I send someone a link to my group order, they can add whatever they want, but ultimately I decide if the order goes through or not. Realized that flow would solve at least half the use cases I have for "shared threads", so it's now built into T3 Connect in T3 Code (nightly)

♥ 1,052⟲ 21💬 129👁 45.2k

T3 Code's sidebar groups threads into collapsible Working, Snoozed and Settled sections, so running threads stay out of sight and only threads needing attention are visible.

T

Theo - t3.gg@theo

View on X ↗

It's so nice to have 14 threads running and zero visible. Why let working threads get in the way of things that need your attention?

♥ 960⟲ 9💬 140👁 44.2k

Superset's PR view now puts a whole PR in one place (every comment, commit, review thread and check); you can reply right on a line and hand the fix to your agent.

K

Kiet@FlyaKiet

View on X ↗

Superset makes code review sexy again 🎷 (sound on) the PR view in @superset_sh now has the whole PR in one place: every comment, commit, review thread and check. reply right on the line, then hand the fix to your agent.

♥ 37⟲ 2💬 10👁 2,480

Paperclip v2026.1023.0: tasks and projects can be made private, you can route agents through your own model gateway (OpenRouter, Bedrock, Gemini), Slack bots set themselves up, and agent costs can be reconciled.

Codex Day 19: changing reasoning effort mid-conversation no longer invalidates the prompt cache, and the Codex CLI adds reusable prompt presets, inline model and permission settings, and enterprise MCP sign-in.

T3 Code iOS 2.1.0 shipped: guided PR reviews on the phone with PRs linked to threads and cached for offline, an offline send queue that holds messages until the environment reconnects, in-thread search over messages and plans, and mobile automations (webhook-triggered tasks with event filters, and agents that watch a PR for checks and review feedback).

J

Julius@jullerino

View on X ↗

T3 Code iOS 2.1.0 is out on the App Store • Pull requests on the go: Browse changes, follow guided reviews, check activity, and link pull requests to threads. Previously loaded pull requests remain available offline. • Built for iPhone Duo and iPad: Improved split layouts, navigation, and transitions, with better continuity when folding, unfolding, or resizing. • A better composer: A compact design that expands as you type, with improved attachments, voice controls, keyboard handling, and consistent behavior between drafts and conversations. • More reliable offline messaging: Queue messages while disconnected and keep track of pending sends until your environment reconnects. • Richer conversations: Search messages and plans within a thread, view LaTeX equations and interactive content, and wrap long code blocks. • More automation options: Create webhook-triggered tasks, filter incoming events, and let agents watch pull requests for checks and review feedback.

♥ 78⟲ 2💬 6👁 1,708

Claude Code CLI now supports MCP Tasks and Elicitations. Tasks let a plugin's MCP server take a long-running job (like deep research) in the background and drop the result back into the session when done; Elicitations let a plugin pause mid-task to ask the user for input or a confirmation.

C

ClaudeDevs@ClaudeDevs

View on X ↗

We're adding support for MCP Tasks and Elicitations in Claude. Tasks let Claude give plugins long-running jobs such as deep research and report back when they're done. / Elicitations let plugins ask users for input mid-task, such as confirming a refund or filling in a missing detail. / Tasks and Elicitations are available today in the Claude Code CLI, rolling out to more Claude products soon.

♥ 1,900⟲ 92💬 118👁 89.7k

Synara Beta 1.0.9 builds a VS Code-style code editor into the app, with TypeScript/JavaScript language smarts (diagnostics, go-to, quick fixes, refactors), plus a built-in JavaScript debugger with breakpoints, stepping, watches and Node launch/attach. It also lets you review a Codex turn's full diff before the turn ends.

Cursor agents can now use Computer Use to click and type across any desktop app, not just the browser, for example driving the iOS Simulator to test what they built. Agents pick it up on their own when a task needs it, or you can @-mention Computer Use in the prompt.

Cursor@cursor_ai

View on X ↗

Cursor can now control your computer. Agents can drive desktop apps like the iOS Simulator to test the features they build. (thread) Computer Use lets agents click and type across any app on your computer, not just the browser. Cursor uses it on its own when a task needs it. You can also ask for it by @-mentioning Computer Use in your prompt.

♥ 1,227⟲ 89💬 132👁 32.2k

Synara v1.0.8 shows running subagents and Devin background jobs above the composer, lets you review live diffs before a turn ends, and makes Hubs keep the original request with delegated work, block duplicate workers on retries, and tell queued follow-ups apart from steering.

Synara@trySynara

View on X ↗

Synara v1.0.8 is out 🪽 🔀 Send, Continue, and Implement now use the same provider handoff; queued messages follow your selected model without silently stopping background work. 🧠 Hubs keep original requests and attachments with delegated work, prevent duplicate workers on retries, and distinguish queued follow-ups from steering. 🤖 Running subagents and Devin background work stay visible above the composer and keep their task identity through recovery. 🧾 Review captured live diffs before a turn ends, including complete Codex patches against the right baseline. ➕ Start a chat in the current workspace from the tab bar; create project actions in Environment and choose which header actions to show. ⌨️ Cmd/Ctrl+W closes the focused chat tab or dock terminal; idle terminals are safer to reuse. 🛟 Codex resume restores saved conversation context and tells you when history is missing or shortened.

♥ 4⟲ 1💬 2👁 135

T3 Code v0.1.2 renders Mermaid diagrams in chat and adds `t3 thread` CLI commands for working with threads from the terminal.

Julius@jullerino

View on X ↗

T3 Code v0.1.2 is out: Mermaid diagrams render in chat via Pele and new `t3 thread` CLI commands. And more importantly bumps Electron and Node.js to block compromised eMudhra CA keys other high severity securtity vulnerabilities.

♥ 144⟲ 2💬 7👁 2,232

Starting next week, Claude's web, desktop and mobile app sessions run in auto mode by default: instead of approving each step, a separate automated check blocks risky actions, and users can switch back to manual anytime. This extends the classifier-based auto mode that became Claude Code's default in August.

Claude@claudeai

View on X ↗

Starting next week, sessions across Claude web, desktop, and mobile apps will run in auto mode by default on Pro, Max, and Team plans. Rather than asking you to approve each step, an automated check blocks risky actions. Switch back to manual anytime.

♥ 3,394⟲ 101💬 197👁 115.3k

Superset Pages (agent-built web interfaces, even 3D) can now be shared with specific people or the whole team. When a viewer leaves a comment on the shared page, the agent reads it and ships the fix.

Kiet@FlyaKiet

View on X ↗

Superset Pages lets your agent ship any web interface to you, even in 3D. We just upgraded its ability to share: You can now share what your agent built with exactly who you want, or the whole team. when someone comments, the agent reads it and ships the fix

♥ 3⟲ 1💬 2👁 119

Factory now takes Slack delegations: type "@Factory fix it" in a Slack thread and Droid checks whether the bug is tracked, writes a failing test, ships the fix and can post a video of it working. For bigger tasks it opens a per-task Slack "code channel" showing the repo, branch and a live diff; anyone in the channel can steer Droid without @-mentioning it, and when done it posts a summary with PR links and archives the channel.

Factory@FactoryAI

View on X ↗

Factory now supports @SlackHQ delegations. Take a task from a conversation to pull request in the thread where your team is already discussing it.

♥ 28⟲ 4💬 6👁 1,492

Paseo 0.13 beta lets plugins replace the built-in file panels or add new ones; the founder says diff and other panels are next, so users can swap in their own diff viewer.

Amp's Orbs, its cloud sandboxes where each agent thread gets its own machine that pauses when idle and resumes on the next message, can now be self-hosted for free by individuals and teams. You run `amp orb provider ./dispatch.sh` in your network. It connects out to Amp over HTTPS only, and your script handles spawn, pause and delete jobs on any compute; there's a ready-made Kubernetes option for GKE, EKS and AKS. Docs: https://ampcode.com/docs/orbs/self-hosted

Thorsten Ball@thorstenball

View on X ↗

You can now self-host orbs. It's free for individuals and teams. https://t.co/FSFcICRPZ0 https://t.co/DSXGjOGC85

♥ 645⟲ 49💬 51👁 135.9k

Zed v1.25: external agents connected over ACP (Agent Client Protocol) now support context compaction and session notices inside Zed's agent panel.

T3 Code now runs agent-written workflows: the agent writes a short script that splits a job into phases (e.g. Map, then Build), and T3 shows a 'Workflow started' card with each phase's subagents, their live status, run times and a viewable script. Julius (core dev) says it shipped last week.

W

Will Undrell@WillUndrll

View on X ↗

T3 code just generated a workflow for subagents to carry out in sequence. When did this ship? @jullerino nice work!!

♥ 6⟲ 0💬 1👁 477

Claude Duet (research preview, waitlist, Pro/Max US): a new app where Claude acts as a proactive agent. It surfaces what needs your attention, does tasks you ask, keeps working when you step away, and checks with you before acting. Uses your Claude memory, skills and connected apps.

C

Claude@claudeai

View on X ↗

We’re opening a research preview of Claude Duet, a new app where Claude works as a proactive agent. It comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf.

♥ 6,604⟲ 483💬 392👁 237.1k

Claude Security (public beta on Claude Enterprise) now re-scans connected repos as changes land after a deep first scan, auto-marks findings fixed or reopens them, writes an editable per-project threat model, and exposes findings via a new API and MCP connector.

Claude@claudeai

View on X ↗

Claude Security now keeps up with your code as it changes. Connect your repositories once. Claude runs a deep first scan, then re-scans as changes land, so findings stay close to the current state of your code. In public beta on Claude Enterprise.

♥ 212⟲ 10💬 37👁 18.8k

Grok Build 1.0.67: agent worktrees reuse the main checkout's saved "always allow" tool approvals (never-allow still applies), so you aren't re-prompted in each worktree; can be turned off in config.

Puck@GrokInsider

View on X ↗

Grok Build CLI just got an update with v1.0.67! Features: • Worktrees also use their main checkout's saved "always allow" answers, so you are not asked again for a tool you already allowed there. Set `[features] worktree_inherits_main_approvals = false` to turn this off; the main checkout's "never allow" answers apply either way.

♥ 0⟲ 0💬 1👁 112

Factory now takes delegations from Jira: assign a task or sub-task to Factory (or mention it in a comment) and its Droid agent investigates, implements and tests it, then reports progress back in Jira. Admins set default model, instructions, plugins and compute once.

Factory@FactoryAI

View on X ↗

Factory now supports @Jira delegations. Delegate tasks and sub-tasks in Jira to Factory, and Droid will start running.

♥ 27⟲ 2💬 5👁 2,113

After you switch Claude Code or Codex accounts, Orca shows a dialog to restart and resume the current session, plus a button to restart and resume all active sessions so they pick up the new account.

Orca@orca_build

View on X ↗

@menjilx Just recently added a button to help restart & resume all the sessions. This way it's more reliable.

♥ 0⟲ 0💬 0👁 12

T3 Code's new Stable release (1,400+ changes) adds EAS Simulator remote hosts for iOS development, PR diffs that split code vs tests vs docs at a glance, early compaction of idle Claude threads to keep the prompt cache warm (beta), filters and dedupe for webhook tasks, and a `t3 connect` reachability check with fix-it errors.

Theo - t3.gg@theo

View on X ↗

We just shipped an updated T3 Code Stable release with over 1,400 changes 🤯 Quick highlight reel of my favorites: - Massive perf wins on desktop and mobile - Add a project to another machine from the "Run on" menu - EAS Simulator remote hosts for iOS dev - PR diffs now show code vs tests vs docs at a glance - Idle Claude threads compact early so the cache stays warm (beta) - Mobile got a big native pass. It finally feels native - `t3 doctor` can reclaim database space - `t3 connect` now checks your machine is reachable, and errors say how to fix them - Webhook tasks got filters and dedupe - Skill discovery is much more reliable - Tons of Windows, terminal, and scroll fixes Least favorite: I caved, LaTeX renders on web and mobile

♥ 261⟲ 3💬 37👁 8,600

OpenCode's iPhone app shows up in the iOS share sheet, so you can send a screenshot straight into a new or existing OpenCode session with a message (Dax sent a DM screenshot and said "schedule this").

dax@thdxr

View on X ↗

wow this flow is amazing, take a screenshot on your phone and send straight to an opencode session i just took one of DMs on X scheduling a call and send it to opencode saying "schedule this"

♥ 44⟲ 1💬 12👁 3,765

Orca's agent chat previews HTML designs inline (e.g. ask for 5 UI options), and you can share the final design with your team as an artifact. Works with Claude Code, Grok, Antigravity, Pi and more.

Orca@orca_build

View on X ↗

🎨 Preview HTML designs right in Orca’s agent chat, then share your final design with the team as an artifact. Works with Claude Code, Grok, Antigravity, Pi, and more. Start with the prompt: “Let’s brainstorm 5 different UI designs in HTML…”

♥ 0⟲ 0💬 0👁 96

Warp released a /merge-risk skill that grades how risky it is to merge a PR across five areas: whether the code matches the PR description and tickets, blast radius and sensitive paths touched, end-to-end verification (screen recordings, before/after screenshots), rollback cost, and whether errors are logged or swallowed. The aim is letting devs trust agents on low-risk changes. Install with `npx skills add warpdotdev/common-skills --skill merge-risk`.

Ben Holmes@BHolmesDev

View on X ↗

We made a skill for assessing the risk of merging a PR: /merge-risk This addresses the code review bottleneck by helping devs trust the agent for lower risk changes. It grades risk across 5 categories: alignment, impact, verification, rollback cost, and error detection.

♥ 914⟲ 35💬 38👁 89.6k

Delta (Zed's agent workspace) has built-in tools to search and read your past threads, so you can ask the agent to find an old conversation just by describing what you remember.

Zed@zeddotdev

View on X ↗

Lost a thread? Ask Delta to find it. Delta has built-in tools to search and read your past threads. Just describe what you remember.

♥ 31⟲ 1💬 5👁 2,244

Claude Desktop 3.1623.0 (bundling Claude Code 2.1.304) adds a Storage view that shows how much disk Claude is using and lets you delete inactive Code sessions and Cowork tasks. Session cleanup now also keeps archived worktrees with uncommitted changes instead of deleting them.

Claude Code Desktop ChangeLOG@ClaudeDesktoLOG

View on X ↗

Claude Desktop 3.1623.0 · Bundled Claude Code 2.1.304 ✨ New • Storage shows Claude's disk use and lets you remove inactive Cowork tasks and Code sessions; not yet available in third-party deployments 🛠️ Fixed in Code • Inactive-session cleanup deleting archived worktrees and uncommitted changes; worktrees changed after preview are kept too

♥ 1⟲ 0💬 0👁 9

Factory now takes Linear delegations: assign a Linear issue to Factory and Droid starts from its title, description, labels and comments, moves it to In Progress, makes and tests the change, opens a PR and links it back on the issue.

Factory@FactoryAI

View on X ↗

Factory now supports @Linear delegations. Delegate an issue in Linear to Factory, and Droid will start running. (thread) Droid starts from the title, description, labels, and comments. Droid moves issues to In Progress, investigates, makes the change, runs the tests, and opens a pull request. Factory attaches the PR link to the issue.

♥ 38⟲ 4💬 5👁 1,239

Superset launched Automation Triggers: an outside event such as a Sentry error, a review request or a bug report in Slack starts an agent run on its own, so a fix PR can be waiting by morning. Works with GitHub, Slack, Linear, Sentry, Notion, Microsoft Teams and webhooks.

Kiet@FlyaKiet

View on X ↗

A production error fires at 2am. By morning, a PR with the fix is waiting for you. Introducing Automation Triggers in @superset_sh: a Sentry error, a review request or a bug report in Slack now starts an agent run on its own. (reply) Works with GitHub, Slack, Linear, Sentry, Notion, Microsoft Teams and webhooks, with more on the way.

♥ 12⟲ 2💬 2👁 429

Hugging Face's Leandro von Werra launched Agent Manager: orchestrate dozens of agents across harnesses (Claude Code, Codex, OpenCode and more) that message each other and self-organize to split up complex tasks, and keep running on Hugging Face Spaces.

Leandro von Werra@lvwerra

View on X ↗

Agent Manager is finally out! Orchestrate dozens of agents across harnesses (Claude Code, Codex, OpenCode etc.) effortlessly. Agents communicate and self-organize to help you achieve complex tasks and always keep running on HF spaces.

♥ 39⟲ 7💬 18👁 5,151

T3 Code can now move a repo you have on one machine over to another of your connected machines in two clicks, so you can keep working on it there. It even carries over .gitignore'd files such as .env if they exist in the repo. The PR is merged and lands in the next nightly.

Theo - t3.gg@theo

View on X ↗

Got a repo on one PC and want to work on it from another? Takes 2 clicks now :) Even handles .gitignore'd files like .env if you have them in the repo

♥ 322⟲ 1💬 42👁 13.6k

T3 Code's PR view now splits a PR's diff stats by category: hovering the +/− totals shows lines added and removed for Code, Tests and Mocks separately. Theo copied the idea from codybrouwers' custom code-review tool within hours.

Theo - t3.gg@theo

View on X ↗

yoink'd straight into T3 Code :)

♥ 120⟲ 3💬 16👁 3,415

Orca's CEO teased Chrome Web Store extensions in Orca's built-in browser: the store page shows an "Add to Orca" button (Dark Reader in the screenshot), so users could install Chrome extensions into Orca's browser.

jinjingliang@JinjingLiang

View on X ↗

Cooking up something 👀 🌐 🫍

♥ 2⟲ 0💬 0👁 41

Devin added an Efficiency Score that estimates the human hours each session saved per unit of spend, plus auto-approvals that give high-scoring engineers more budget automatically while other requests go to an admin with reason, efficiency and model mix. Devin Coach helps everyone else improve.

Cognition@cognition

View on X ↗

What if your AI budget went to the engineers who use it best? Blanket usage caps are a hiring freeze in disguise. Introducing efficiency-based auto-approvals for your most productive engineers, plus Devin Coach to help everyone get there. (thread) We created an Efficiency Score to show how much real engineering work each person gets out of Devin. For every session, including investigations and planning that never ship a PR, it estimates how long a human engineer would have needed, then scores productive hours per ACU. In Devin, engineers with a healthy grade can get more automatically. Other requests go to an admin, who sees their reason, efficiency, and model mix before approving.

♥ 105⟲ 10💬 11👁 7,851

Superset launched Plugins: install an MCP server, skill or login once and every agent in Superset (Claude Code, Codex, Cursor, or your own) gets it on every machine, so you stop configuring the same tools per agent. Users said it fixes re-setting up MCPs in remote workspaces. Kiet says org-wide plugins are coming this week.

Kiet@FlyaKiet

View on X ↗

Every few months the best coding agent changes. Your tools shouldn't have to. Today we're launching Plugins in @superset_sh. Install once and every agent gets it, on every machine. No more setting up the same MCPs, skills and logins for Claude Code, Codex and Cursor. (reply) Works for any agent running in @superset_sh. you can even build your own

♥ 8⟲ 1💬 3👁 185

Traycer's desktop composer now has voice input: hit the mic or Ctrl+Shift+M and talk, and the speech is transcribed on your device and lands in the composer as you pause. It's pitched for long, rambly instructions.

T

Traycer@TraycerAI

View on X ↗

some prompts are faster to say than type. hit the mic or ⌃⇧M and talk. speech is transcribed on your device and lands in the composer as you pause. made for the long, rambly instructions.

♥ 2⟲ 0💬 0👁 165

Orca's mobile app now uses the same sidebar filter as the desktop app, part of a push to bring desktop capabilities to mobile, and Orca says more improvements to the mobile PR review flow are on the way.

Orca@orca_build

View on X ↗

We’re bringing more desktop capabilities to the mobile app 📲 In the latest version, the mobile app now share the same sidebar filter as the desktop app. More improvements to the mobile PR review flow is on the way too.

♥ 14⟲ 0💬 1👁 544

T3 Code has an open PR (#19662) adding an opt-in "Compact idle threads" setting: a long idle thread is compacted just before the provider's prompt cache expires, so coming back only reprocesses the summary, not the whole context. Maria is asking whether to ship it across all providers.

maria@maria_rcks

View on X ↗

should we? (across all providers) [screenshot: pingdotgg/t3code PR #19662 "feat(server): compact idle threads before the prompt cache expires" — opt-in setting Compact idle threads (Settings > General); Claude Code shipped the same behavior]

♥ 43⟲ 0💬 15👁 903

Kit Langton (OpenCode team) demoed Scrawl, a macOS app for briefing coding agents: press a shortcut, talk while you click and draw on screen, then paste one bundle into the agent chat. It holds a timestamped transcript with numbered click and drawing tags, plus one image grid cropped to just the spots you touched, instead of a video or full screenshots. Shown pasting into OpenCode; repo still private.

Kit Langton@kitlangton

View on X ↗

hey. here's a little demo of the most meaningful improvement to my productivity since speech-to-text.

♥ 99⟲ 6💬 13👁 2,727

Anthropic published two official Claude Code plugins: Receipts (/receipts week) summarizes sessions, files touched, commits and PRs shipped and each project's share of usage as a printable HTML receipt plus CSV, and Session Report breaks down tokens, cache efficiency, subagents, skills and the most expensive prompts in a clickable HTML report.

0xMarioNawfal@RoundtableSpace

View on X ↗

Anthropic made two official Claude Code plugins that show exactly what your usage bought you. Receipts prints a receipt of your Claude Code work: → Sessions, files touched, commits and PRs shipped → Each project's share of your usage → An HTML receipt you can print to PDF, plus a CSV export /plugin install receipts@claude-plugins-official Then: /receipts week Session Report shows where your tokens went: → Tokens, cache efficiency, subagents and skills → Your most expensive prompts, in an HTML report you can click through /plugin install session-report@claude-plugins-official Both read your local Claude Code history. Receipts sends only counts and project names to write the report, and the report stays on your machine.

♥ 11⟲ 0💬 4👁 15.5k

Theo confirms T3 Code supports cross-harness subagents: a Claude session can spawn Codex subagents, so one agent delegates work to another vendor's agent inside the same thread.

Theo - t3.gg@theo

View on X ↗

@ThisMightWrk Claude can have codex subagents in t3 code :)

♥ 0⟲ 0💬 0👁 113

Sneak peek of iOS Live Activities for Synara Remote: lock-screen widget shows the running agent task, connected Mac, and project folder with a Running spinner. Parenthetical also previews embedding VS Code Code OSS as Synara's in-app editor.

Emanuele Di Pietro@emanueledpt

View on X ↗

Sneak peek at the Live Activities in the Synara iOS app (adding VSCode Code OSS to power the editor inside Synara)

♥ 8⟲ 2💬 0👁 354

JetBrains' open-source ADE ThinkRail (Apache 2.0 incubator, built on the pi agent) keeps a linked graph of SPEC.md files beside the code that the agent reads and updates through dedicated spec tools, so project intent survives across sessions. It also has a worktree per workspace that auto-shelves when its PR merges, and review comments anchored on files and diffs that go to the chat as structured context (confirmed in the JetBrains/thinkrail README).

小墨同学@xiaomovps

View on X ↗

JetBrains开源的 Pi 桌面端 ThinkRail，大厂实力强啊🔥 我看了一下这个项目基座是：pi-coding-agent，但是基于这个项目 他还做额外的优化，比如说针对代码方面 JetBrains 就发挥了他的强项 优点： 1. 底座就是 pi-coding-agent 模型、技能、会话还是 Pi 自己管 2. 每个工作区一个 git worktree 并行不碰你手上的代码 3. 内置编辑器和 Git 变更视图 能在 diff 上留评论直接发给 Agent 4. 用 SPEC.md 维护项目规格 上下文跨会话留存 社区给 Pi 套 GUI 的不少，但是这次是 JetBrains 亲自下场 Apache 2.0 开源 没有自己的账号，也可以直接走 Pi 的认证 现有技能和 mcp.json 接着用，无缝的衔接直接去使用，非常建议大家去使用

♥ 39⟲ 5💬 4👁 3,596

T3 Code can load-balance new threads across your connected machines: turn on Settings → Connections → Load balancing and it picks the machine with free CPU and memory, with per-machine Prefer / Less often / Manual only settings (confirmed in T3's remote-access docs). A user running 3 Macs and a phone says it shifts work when one machine's CPU gets busy.

Pierre-Olivier Carles@pocarles

View on X ↗

T3 Code is really good. I finally understand now how @theo works on so many devices at the same time. I have 3 Macs and 1 phone, all in sync and there's even load balancing when one's CPU gets bad. Really good job, @jullerino and team. PS: And the Dark Ember skin is stunning.

♥ 1⟲ 0💬 0👁 67

Traycer tasks can now span several repos: attach the API and web app folders to one task so its agents work across both, and each repo runs either in place or in its own worktree.

Traycer@TraycerAI

View on X ↗

the change spans the api and the web app. so should the task. attach both folders to one task so its agents can work against each. every repo gets its own run location: in place, or in its own worktree.

♥ 0⟲ 0💬 0👁 66

Cursor's Environments API (beta) lets you create and manage cloud agent environments in code, including which repos they clone and the environment.json setup they run. Environments now hold up to 500 repos and big multi-repo ones start 5x faster.

Cursor@cursor_ai

View on X ↗

The Environments API is now available in beta. Create and manage cloud agent environments programmatically, including the repos they clone and the environment.json setup they run.

♥ 8⟲ 0💬 3👁 1,124

Cursor cloud agents can now build interactive canvases: ask for a dashboard or report while the agent works, then share the finished canvas with your team as a link.

Cursor@cursor_ai

View on X ↗

Cloud agents can now build interactive canvases. Ask for a dashboard or report while the agent works, then share the finished canvas with your team as a link.

♥ 77⟲ 7💬 17👁 4,372

Amp lets you see and control your macOS or Linux runner's desktop from the Amp client (browser or phone): turn on Share This Mac's Display under Settings › Runner, or start with `amp --no-tui --desktop`. Useful for trying native apps Amp builds (e.g. iOS apps) with a near-instant local-network connection.

Thorsten Ball@thorstenball

View on X ↗

You can now share your Amp runner's desktop https://ampcode.com/news/runner-desktop

♥ 44⟲ 2💬 1👁 2,217

Delta (Zed's agent workspace) can configure its own settings from the chat: in the demo the user asks it to add Ollama as a custom OpenAI-compatible provider and never opens Settings.

Zed@zeddotdev

View on X ↗

In this video, Joseph asks Delta to configure itself to use @ollama as a custom OpenAI-compatible provider. It all happens in the thread, without Joseph ever touching Settings.

♥ 23⟲ 0💬 5👁 2,286

Cline Desktop now starts Cloud Agents straight from the desktop app: run several tasks in parallel, each in its own cloud sandbox, close the laptop while they keep working, and check in from your phone.

Cline@cline

View on X ↗

Cline Desktop now runs in the cloud! ☁️ Run multiple tasks in parallel - each in its own secure cloud sandbox - and close your laptop while Cline keeps working. These are the same Cloud Agents you can run from your browser, now built into the desktop app. Start a session on desktop and check in on it from your phone.

♥ 18⟲ 2💬 4👁 1,080

Devin can now build and test foldable iPhone Duo apps in macOS cloud VMs: it unfolds, folds and closes the virtual phone, checks every screen, and sends video verification of what it tested.

nader dabit@dabit3

View on X ↗

Devin can now build and test iPhone Duo apps in macOS cloud VMs Unfold the phone, fold it, close it, check every screen, and get video verification of everything. Here are some examples: 1. Devin takes someone's portrait while they watch themselves on the outside screen, it sends posing cues, half-folds the phone so it stands on a table, takes timer shots, and closes the phone mid-countdown to make sure no photo gets lost

♥ 66⟲ 11💬 15👁 5,137

Factory Delegations is now generally available: anyone on a team can hand a task to the Droid agent from the tools they already use. An admin sets the default model, instructions, plugins, compute and a dedicated service account (with its own Git credentials and connectors) once in Settings, and Droid then picks up delegated work on a ready remote machine and opens PRs as a bot.

Factory@FactoryAI

View on X ↗

Introducing Delegations, now generally available. Your team can hand work to Droid from the tools they already use. (thread) An admin sets the default model, custom instructions, plugins, service account, and compute once, in Settings > Delegations. After that, anyone in the org can delegate a task, and Droid picks it up on a remote computer that's ready to go. Shared work runs under a dedicated service account, with its own Git credentials and connectors, not as whoever made the request. Admins decide what Droid can access. Delegations are the foundation for bringing Factory into the tools your teams already use. More on that next week.

♥ 28⟲ 2💬 2👁 1,028

Paperclip added you.com as a connector: agent teams get web search and page extraction inside the same project (researcher can hand a sourced brief to a writer), with searches landing in Paperclip's audit log; covers Claude, Codex, Gemini, Cursor and other runtimes.

Paperclip@papercliping

View on X ↗

Your agent team needs current sources, not another tab for you to babysit. @youdotcom is now in Paperclip. Give your researcher web search and page extraction, hand a sourced brief to your writer, and review the result. All within the same project.

♥ 3⟲ 1💬 1👁 119

Traycer lets you review a plan before the agent writes code: highlight a line in the spec, leave a comment like a PR review, then ask the agent to work through your comments and resolve the threads as it goes.

Traycer@TraycerAI

View on X ↗

review the plan before the agent writes the code. highlight a line in the spec and comment on it, like a PR review. then ask the agent to work through your comments, and it can resolve threads as it goes. https://t.co/Ndz5HyPFpc https://t.co/UOm3YM3Hw5

♥ 1⟲ 0💬 0👁 120

T3 Code's new stable release (with Orchestrator V2) includes a community PR that gives agents worktree tools: an agent can hand its worktree off to another session and report its state, so T3 Code tracks which session owns each worktree and parallel agents pass work along without colliding (community contributor @nsxdavid; low traction so far).

nsxdavid@nsxdavid

View on X ↗

T3Code Stable with Ov2 has landed! Which also means, my PR has finally landed. Why would you care... we... My little addition to this project PR gives T3 Code's agents tools for working in git worktrees, the separate checkouts each agent works in. An agent can hand its worktree off to another session, and it can report what state that worktree is in. That lets orchestrated parallel agents pass work along and keep track of it without stepping on each other. Without this tool, an agent can still run git, but git only knows about files, not about T3 Code. The agent can't tell which session owns a worktree, whether another agent is mid-task there, or how to hand its work to a new session so T3 Code's interface and orchestrator follow along. Agents end up guessing from raw git output, colliding in the same checkout, or leaving worktrees orphaned. With this tool, T3 Code itself tracks who owns each worktree and what state it's in, so handoffs between parallel agents are deliberate and visible instead of improvised with shell commands. You're welcome.

♥ 1⟲ 0💬 1👁 14

Traycer now lets you tile tabs into split panes: drag a tab to the edge of a group to split it or into the middle to join it, or press ⌘D and pick what goes in the new pane, so the agent chat, its terminal and the spec sit on one screen.

Traycer@TraycerAI

View on X ↗

your agent, its terminal and the spec, all on one screen. drag any tab to the edge of a group to split it, or into the middle to join it. tile it the way you actually work. (follow-up: prefer the keyboard? ⌘D splits the group, then pick what goes in the new pane, like a terminal next to the chat.)

♥ 3⟲ 0💬 1👁 147

Dax (OpenCode co-founder) showed the OpenCode phone app driving the OpenCode server on his office computer while the same session works in a team Slack thread through the OpenCode Slack bot, posting updates back to the thread (199 likes, 7k views).

dax@thdxr

View on X ↗

i am walking outside using the opencode app on my phone it controls the opencode server on my computer in my office and i'm having it work in a thread in slack with my team and our opencode slackbot

♥ 199⟲ 3💬 39👁 7,309

Cognition launched Devin Oncall: Devin automatically picks up production alerts and bug reports from Datadog, Dynatrace, Sentry and similar tools, investigates them and helps diagnose and fix incidents (255 likes, 13k views).

Cognition@cognition

View on X ↗

Introducing Devin Oncall: Devin automatically investigates alerts and bugs, helping your team diagnose and resolve production incidents. Devin Oncall already handles thousands of alerts every day for some of the world’s largest companies. Let it become your SRE today.

♥ 255⟲ 25💬 23👁 13.1k

Agents can now drive Superset's own UI: through the Superset CLI and skills, an agent can control and script how panes, sessions and views are laid out and displayed. The demo shows an agent rearranging the app in real time with no human input.

Superset@superset_sh

View on X ↗

Your agent can now drive Superset UI It can control and script how everything displays. Available through our CLI and skills 🪄 This demo is in real-time, completely agent driven. Sound on 🔉

♥ 12⟲ 1💬 1👁 434

v1.0.5 ships the previewed in-chat visuals (interactive charts, tables, checklists, calculators; sandboxed HTML visuals on desktop) and adds iOS Simulator tools so agents can launch apps, inspect screens, read logs and set up tests. It also adds an Activity view that puts approvals, questions and ready plans first, with ⌘⌥A to jump to the next chat that needs you, groups subagent chats under their parent, and adds a compact composer with shortcuts for tasks, goals and changes.

Synara@trySynara

View on X ↗

Synara v1.0.5 is out 🪽 Your agents can now create visuals, surface what needs you, and test iOS apps. 📊 Interactive charts, tables, checklists and calculators in chat; sandboxed HTML visuals on desktop. ✍️ A compact composer that expands as you write, with floating shortcuts for tasks, goals and changes. 🚦 Activity puts approvals, questions and ready plans first; ⌘⌥A / Ctrl+Alt+A jumps to the next chat needing you. 🤖 Subagent chats stay grouped with their parent instead of crowding the sidebar. 📱 iOS Simulator tools can launch apps, inspect screens, read logs and configure tests—with Xcode 27 input fixed. 🧠 Claude usage totals no longer double-count resumed sessions; Inbox separates cache reads. 🖱️ Browser agents keep your composer focused: 0 focus losses across 2,400 test hovers. ⚡ Task-heavy chats avoid repeated database scans; Claude settings changes preserve resumable sessions. 🎨 A new animated ASCII chat landing, with a still option for reduced motion.

♥ 5⟲ 3💬 2👁 94

Claude Code 2.1.299 added warnings to /usage that point out what is driving usage up: max effort, workflows, always-on context and large tool results. Confirmed in the official Claude Code CHANGELOG (2.1.299 entry).

ひろ｜一人社長のClaude Code導入支援@hiro_hitori_ai

View on X ↗

Claude Code v2.1.299、/usage に警告表示を追加 警告が出るのは、max effort（最大の処理設定）とworkflows（作業手順）。 さらにalways-on context（常時読み込む情報）と大きなツール結果も対象です。 出典: Claude Code 公式CHANGELOG

♥ 0⟲ 0💬 0👁 9

Claude Code desktop: when you ask Claude to run tasks in parallel it starts a nested session for each one, and each reports its result back into the main session so the thread stays readable. Claude also suggests follow-up work while it runs that you can launch as its own session in one click, and /btw opens side questions in their own tab.

ClaudeDevs@ClaudeDevs

View on X ↗

When you ask Claude to run tasks in parallel, it starts a nested session for each one. Each reports its result back into your main session when it's done, so the main thread stays readable. https://t.co/6gxVlCT1e4 Claude also proactively suggests follow-up work as it's working, and you can run that as its own session with just one click. You can also use /btw, so side questions open in their own tab while the main session keeps going. https://t.co/ytGlqoyEi1

♥ 247⟲ 4💬 5👁 10.6k

Claude Code desktop is rebuilt for running many sessions side by side: a sidebar shows which sessions are working and which need you, each session has a live to-do list that follows Claude's plan, a recap card when it needs you, and a side rail with where it runs, its context and its changes. You can also save setups (folders, an SSH host or the cloud, optional worktree per folder) as environments to start sessions from.

ClaudeDevs@ClaudeDevs

View on X ↗

We've been working hard on making the Claude Code desktop app great for running lots of sessions side by side. It's now much quicker to start sessions, easier to split work across a few, and you can always see what each of them is doing. Here’s what's new: https://t.co/iF1WuEueMB The sidebar makes it easy to see which sessions are working and which ones need you. Inside a session, a live to-do list follows Claude's plan as it goes, a recap card sums up the turn when Claude needs you, and the side rail shows where it runs, its context and its changes. https://t.co/uJ5zBJGbZu It's now much easier to pick up where you left off. The new session screen keeps your recent folders one click away. You can also save a setup as an environment, like a few folders, an SSH host or the cloud, and start sessions there, with the option of a worktree per folder. https://t.co/kybTW1YVdX

♥ 3,714⟲ 111💬 181👁 112.1k

Amp added Calling Puck: real-time voice calls (powered by GPT-Live-1) with Puck, its built-in agent manager, so you can talk to it from anywhere, even a watch, to launch, coordinate and check on coding agents while away from your computer. Amp also launched Sign in with ChatGPT, so users can run Amp on their ChatGPT plan usage (including GPT-6 Astra). Source: OpenAI's Oct 14 customer story (openai.com/index/amp-beyond-coding).

RJ@RJiOSDEV

View on X ↗

Amp says some users now burn through 10 billion AI tokens a week. On Oct 14, OpenAI detailed Amp’s “Sign in with ChatGPT” integration, plus a voice assistant that can launch and manage coding agents—even from a watch. Are agent workflows outgrowing the IDE?

♥ 1⟲ 0💬 0👁 23

Traycer wireframes: ask an agent to add a wireframe to a spec and it renders as a live HTML preview inside the artifact; you can open it fullscreen and copy the HTML once the layout is right.

Traycer@TraycerAI

View on X ↗

argue about the layout before anyone writes the css. ask an agent to add a wireframe to the spec and it renders as a live html preview right in the artifact. go fullscreen, then copy the html when it's right. https://docs.traycer.ai/panels/artifacts/wireframes

♥ 2⟲ 0💬 0👁 72

Zed v1.24: drag a Terminal tab into the Agent Panel and it becomes a Terminal Thread, keeping the same running session, title and state, so a plain terminal turns into an agent thread without restarting.

Zed@zeddotdev

View on X ↗

You can now drag Terminal tabs into the Agent Panel to use them as Terminal Threads, keeping the same session, title, and state. Thanks cppcoffee!

♥ 2⟲ 0💬 1👁 221

Factory launched OpenCompute (private preview): a self-hosted orchestrator for agent sandboxes that you deploy once in your own VPC on the runtime you choose. It spins sandbox machines up when sessions, scheduled Automations or incident investigations need them and shuts them down when they finish, so a whole organisation's fleet of cloud agents runs on its own infrastructure.

Factory@FactoryAI

View on X ↗

Introducing OpenCompute: a self-hosted orchestrator for agent sandboxes on your choice of runtime. Adopting autonomous software development should not require handing over your valuable code and context to someone else’s cloud. Scale your fleet of remote agents while keeping them in your VPC. (thread) Meet the demand for cloud agents running across entire organizations. Enable interactive sessions as well as agents operating independently as work arises. Deploy OpenCompute once into your environment, and let it handle orchestration across your entire fleet of agent sandboxes. Scalable cloud agents, securely in your environment, with zero infrastructure hassle. OpenCompute handles the sandbox lifecycle for you: spinning new machines up as they are needed, and down once sessions are over. OpenCompute is now available in private preview.

♥ 88⟲ 9💬 8👁 2,619

AgentPlay (free, local, open source, from the Entelligence founder) turns existing Claude Code, Codex, Cursor and Copilot sessions into interactive maps and replayable timelines. You can see what the agent explored, where it failed, what it retried and how it reached the final change. It drew 57 likes, 38 replies and 2.6k views in its first 35 minutes.

Aiswarya Sankar@Aiswarya_Sankar

View on X ↗

Introducing AgentPlay 🚀 The future of coding is self improving coding agent harnesses that help agents avoid making the same mistakes over and over. Turn your existing Claude Code, Codex, Cursor and Copilot sessions into interactive maps and replayable timelines. See what your agent explored, where it failed, what it retried and how it reached the final change Free, local and open source. GitHub: https://t.co/QR9aGmRy1C

♥ 57⟲ 12💬 38👁 2,568

Slack Code (Salesforce demo): a Slack code channel turns a bug report into a shipped fix in one shared thread, with the whole team and coding agents (Claude Code, GitHub Copilot, Devin) planning, building and reviewing together in the open instead of one person prompting privately.

Salesforce@salesforce

View on X ↗

.@SlackHQ just unlocked multiplayer mode—with Slack Code Watch Katie Steigman show how a code channel turns a bug report into a shipped fix — in real time, with the whole team and their AI agents working the same thread. That's what multiplayer AI looks like: not one person prompting in private, but people and agents like Claude Code, GitHub Copilot, and Devin building together, in the open.

♥ 9⟲ 1💬 1👁 1,301

Orca worktrees can now hold a team of nested worktrees: a manager agent in the parent spins up isolated child worktrees, each running its own agent loop, and oversees them so the work lands in order. Coming soon: nested worktrees across repos, so one manager can coordinate work in several repos.

Orca ADE@orca_build

View on X ↗

An Orca worktree can contain a whole team of worktrees 🤯 A manager agent in the parent spins up isolated nested worktrees, each running its own agent loop. It oversees the team and makes sure everything lands in order. Coming soon: nested worktrees across repos—so one manager https://t.co/FVGz8XkbBT

♥ 28⟲ 1💬 3👁 932

Synara MCP Connectors: connect once (Supabase, Sentry, Linear, Notion, Cloudflare, Vercel, custom MCP by URL) and every agent in the thread can use them without leaving chat. Now out in Synara Beta 1.0.6-beta.1 (Oct 16) for Claude, Codex and 8 more harnesses via Settings → Connectors; stable soon.

Synara@trySynara

View on X ↗

MCP Connectors are coming to Synara in beta with the next update. Your agent can now query Supabase, triage Sentry errors and update Linear issues without you leaving the thread, and more. Missing one? Add any custom MCP server by URL. https://t.co/DvlviOC2UY

♥ 8⟲ 0💬 0👁 305

Claude Code 2.1.299 makes idle compaction official: when a chat over 200K tokens sits idle for about 54 minutes, Claude Code compacts it just before the one-hour prompt cache expires, so the next message doesn't reprocess the whole conversation. It can be turned off with "idleCompaction": false, and the token threshold is adjustable.

Claude Code Changelog@ClaudeCodeLog

View on X ↗

Claude Code 2.1.299 has been released. 190 CLI changes, 1 system prompt change Highlights: • Added --disallow-tools to claude plugin eval to block specified built-in tools during runs and avoid tool bias • Model accepts security-testing/dual-use prompts and uses AskUserQuestion to clarify, so they are not refused • Idle compaction compacts chats over 200K tokens after ≈54m idle, preventing full reprocessing on resume Full details are in thread below

♥ 33⟲ 3💬 2👁 2,906

v1.0.4 lets Synara's Computer Use paste workspace images and files straight into macOS apps, and makes chats that died in a crash resume on their own after checking what work already exists. It also keeps browser tabs loaded across switches, adds a sidebar sorted by working and unread chats, and pins Hub coordinators with expandable workers.

Synara@trySynara

View on X ↗

Synara v1.0.4 is out 🪽 🖼️ Computer Use now pastes workspace images and files into macOS apps. 🖱️ Computer Use gets more reliable targeting, typing, scrolling and expired-approval recovery. 🌐 Browser tabs stay loaded through ordinary switches, keeping forms, scroll and history. 💬 Full chat history is back, with smoother switching and fewer scroll jumps. ↕️ Your sidebar, your order: chronological activity or working and unread chats first. ♻️ Eligible chats resume after crashes, checking existing work before continuing. 🛡️ Active sessions stay better protected from idle cleanup, stalls and late completion events. 🏡 Easier Hub navigation: pinned coordinators, expandable workers and consistent permissions. 🧵 Background results stay with their reply; completed work stays compact and expandable. 🏷️ Chat and worktree names follow your first message; manual titles stay yours.

♥ 0⟲ 0💬 2👁 39

Amp shipped an MCP server for Puck, its agent manager ("Take Puck with You" on ampcode.com/news): from ChatGPT, Claude, Codex, Grok Bot or any other MCP client you can talk to the same Puck, list your orbs, launch agents and pull results back, so work can be delegated to Amp from other chat apps.

Igal Tabachnik@hmemcpy

View on X ↗

I was just thinking that I needed to connect my ChatGPT/Codex app to @AmpCode and delegate tasks to it, and they just shipped the MCP! https://ampcode.com/news/take-puck-with-you

♥ 1⟲ 0💬 1👁 32

OpenHands now integrates with DigitalOcean Managed Agents: coding agents run in isolated DigitalOcean cloud environments, sessions can be paused and resumed, and everything is managed from Agent Canvas. DigitalOcean can be used for agent execution, model inference, or both.

OpenHands@OpenHandsDev

View on X ↗

OpenHands now integrates with @DigitalOcean Managed Agents! 🌊 Run AI coding agents in isolated cloud environments, pause and resume sessions, and manage everything from Agent Canvas. Use DigitalOcean for agent execution, model inference, or both. More flexibility in where

♥ 2⟲ 0💬 1👁 352

Origin, Cursor's code hosting platform, now connects repos to 20+ integrations for CI, deployments, code review, security, monitoring and project management, and its app marketplace is open to public submissions so any devtool team can publish an app.

Cursor@cursor_ai

View on X ↗

Origin, Cursor’s code hosting platform, now supports over 20 integrations. Connect your repos to tools for CI, deployments, code review, security, monitoring, and project management. (thread) The marketplace is now open to public submissions, so any team making developer tools can publish an app.

♥ 315⟲ 34💬 33👁 12.2k

Cognition launched Devin in Pylon: Devin syncs with the Pylon customer-support platform so support staff and engineers work with the same agent from a customer's bug report through to a merged fix PR, without copying context between tools.

Cognition@cognition

View on X ↗

Introducing Devin in Pylon: Accelerate your Median Time to Fix (MTTF), with $15,000 in credits for new Devin customers. (thread) With our new integration, Devin and Pylon stay in sync so CSMs and engineers can collaborate with the same agent from bug report to a PR merged with the fix.

♥ 90⟲ 8💬 6👁 6,694

Factory launched Remote Control: you connect any machine you manage (a cloud VM, your workstation, a spare Mac mini) to Factory, and its Droid agents run there persistently, scoped to one user or a service, using that machine's repos, services and credentials and keeping durable context on the device across sessions.

Factory@FactoryAI

View on X ↗

Today we are introducing Remote Control. Your Droids run on a persistent self-managed remote machine, scoped to an individual user or service. (thread) Remote Control lets you run Droids on any machine (a cloud VM, your workstation, that old Mac Mini you bought back in November), connect it to Factory, and work with it remotely. Droids on remote controlled machines can seamlessly work with repositories, dependencies, and services, take advantage of any tokens or authentication you provide it, and read and write durable context to the device that can last across sessions.

♥ 59⟲ 5💬 3👁 2,237

Traycer agents can now delegate side quests to child agents, each with its own session and model, shown nested under the agent that started them so you can see who delegated what.

Traycer@TraycerAI

View on X ↗

one agent can't do everything. so let it delegate. agents in traycer spin up child agents for the side quests, each with its own session and model, nested under the agent that started it. you always see who delegated what.

♥ 1⟲ 0💬 0👁 126

Community-built Turn Inspector mod for Claude Code (from claude-code-templates): it shows exactly what's in the context window at every turn, including system prompt, tools, memory, earlier turns, model output and tool results, so you can see where your session's tokens go. It got 1k views and 22 bookmarks in its first hour.

Daniel San@dani_avila7

View on X ↗

Introducing Turn Inspector for Claude Code! A new Mod that lets you inspect exactly what's inside Claude's context window at every turn - System prompt - Tools - Memory - Previous turns - Model output - Tool results Watch Claude's context window fill up turn by turn and finally understand where all those tokens in your session are going Install: npx claude-code-templates@latest --mod ui/turn-inspector

♥ 23⟲ 3💬 6👁 1,026

Unstoppable.ai's desktop agentic IDE runs Claude and Codex together in cross-agent code review loops: a "Code Review (Multi-Provider)" pipeline fans out six reviewers in parallel (Codex, Claude, Opus, GPT) then synthesizes a verdict, shown as a flow graph with run summary and recommendations. The sidebar has New task, Search, Advisor, Automations, Marketplace, Pipelines, Remote Control, plus projects and worktrees with PR badges; an Activity panel shows summary, changed files and PR checks.

Daniel San@dani_avila7

View on X ↗

Interesting approach from @unstoppableweb Their IDE lets you run Claude and Codex together in code review loops, customize multi-agent workflows, and switch between agents in one place The cross-agent code review is what caught my attention https://t.co/vQsU3NGllZ

♥ 51⟲ 2💬 13👁 4,089

Paperclip v2026.1012.0: outside assistants like Claude Code, Codex and any MCP client can now join your Paperclip team as members, navigation is now chat-first with tasks shown beside the chat, and Composio, Arcade and native account connections share one integration catalog. Also adds Claude Sonnet 5.5 and GPT-6.1 Sol.

Paperclip@papercliping

View on X ↗

Paperclip v2026.1012.0 is out 🤝 Claude Code, Codex & any MCP client join your team 💬 Chat-first navigation, tasks beside it 🔌 One catalog for Composio, Arcade & native accounts 🧠 Claude Sonnet 5.5 + GPT-6.1 Sol 130 commits from 9 contributors https://github.com/paperclipai/paperclip/releases/tag/v2026.1012.0

♥ 8⟲ 0💬 2👁 408

Factory made Environments generally available: reusable cloud sandbox definitions (repos, tools, dependencies) that its Droid agents run in, so long-running Missions can fan tasks out across many sandboxes and automations run in a consistent setup while you're logged off. It's the first of a week of cloud-agent launches.

Factory@FactoryAI

View on X ↗

This week we're shipping our vision for the best cloud agent experience. We’re starting with an often-overlooked but critical foundation for any agent: the sandbox where it runs. Today, we’re introducing Environments, now generally available.

♥ 99⟲ 4💬 12👁 4,821

T3 Code's mobile app now lets you drag to set a preferred microphone order for voice input (built-in, wired, Bluetooth, CarPlay), and it records from the first connected mic on the list. Specific wired and Bluetooth devices get added by name after you use them.

Julius@jullerino

View on X ↗

Another mobile improvement, select preferred microphone order

♥ 122⟲ 1💬 12👁 2,312

T3 Code's core dev says voice dictation will get two new options: transcribing on your own environment (the host machine) with a local model like Parakeet, which handles non-English speech better than on-device dictation, and bring-your-own-key cloud transcription endpoints, since the AI labs don't allow their transcription APIs over a subscription.

Julius@jullerino

View on X ↗

Dictating in English or? I’ve found the on device model to work very well, but falls short for other languages. We’re gonna add so you can transcribe on the environment w/ like Parakeet which handles the latter a bit better but English is a wash from my testing And the BYOK for cloud endpoints too. The labs don’t let us use their transcription endpoints over your sub

♥ 6⟲ 0💬 3👁 445

Synara's founder is building Synara Profiles: a public page you can claim that shows your agent work (tokens used, agents run, streak and shipped work), since GitHub graphs no longer reflect agent-driven coding. Synara also says Remote Connections and an iOS/iPadOS app are coming soon.

Emanuele Di Pietro@emanueledpt

View on X ↗

Your GitHub graph doesn’t show how you code anymore Most of my work runs through agents now and nothing shows it So I’m building Synara Profiles, tokens, agents, streak, shipped work, all in one page you can claim

♥ 7⟲ 0💬 4👁 697

Theo confirmed ts-rust (tsc-rs, his agent-built Rust rewrite of the TypeScript compiler/type checker/LSP) is now shipped into T3 Code after an agent in the wrong thread merged it; it ran with no issues for 8 hours so they are keeping it.

Theo - t3.gg@theo

View on X ↗

Oh man, accidentally told the wrong thread to "get this merged" and we shipped ts-rust to T3 Code. No issues for 8 hours...guess we're keeping it?

♥ 64⟲ 1💬 20👁 3,113

Theo shipped an "update everything" button in the latest T3 Code nightly: one click updates the local app, T3 Code on all remote environments, all providers (Claude Code, Codex, etc.) across those environments, then resumes your threads when done.

Theo - t3.gg@theo

View on X ↗

Latest T3 Code nightly now has an "update everything" button. 1 click to: - Update your local app - Update T3 Code on all environments - Update all the providers (claude code, codex etc) across your environments - Resume your threads when done

♥ 217⟲ 2💬 41👁 5,252

Maria (T3 team) says they are exploring native extension/plugin support for T3 Code so people can maintain their own "forks" instead of patch systems, with agent-editable hot-reloading UI. Explicitly not coming soon — they want the implementation right first.

maria@maria_rcks

View on X ↗

so... a bit more context on this one i've been seeing more and more 'extensions', patch systems, and forks (of course) built around t3code lately also a ton of big features people want, but don't make sense to maintain on the repo, or serve a very specific purpose we (and by we i mean mostly me and opus) are exploring what native extension or plugin support could look like in t3code, so yall could maintain your own 'forks' THIS IS NOT COMING SOON, I REPEAT, THIS IS NOT COMING SOON. we want to get the implementation right, not rush something out

♥ 46⟲ 0💬 12👁 2,069

Theo ruled out native LaTeX in T3 Code for good, citing markdown breakage and render cost; he might later add a dedicated math tool so agents can emit LaTeX only for specific reply parts.

Theo - t3.gg@theo

View on X ↗

This particular reply has pushed me over the edge. I have finally decided: there will never be native LaTeX support in T3 Code.

♥ 106⟲ 1💬 21👁 8,053

Theo added a proper /goal flow to T3 Code nightly, where you set a standing objective the agent keeps working toward and the thread shows it as a visual callout (120 likes, 5.4k views). A reply citing T3's docs says the behavior differs by provider: Stop pauses a Codex goal, while Claude keeps the goal and rechecks it after your next message. About an hour later Theo posted a follow-up showing a /goal run that had worked autonomously for over an hour, editing 33 files and finishing a benchmark demo, saying it "wasn't supposed to work, much less in an hour" (https://x.com/theo/status/2109512367403089956).

Theo - t3.gg@theo

View on X ↗

Forgot to mention that I finally caved and added a proper "/goal" flow + visual callout in the T3 Code nightly

♥ 120⟲ 2💬 29👁 5,409

Day 7 of Tibo's Codex ship series: GPT-6.1 Sol's default context in Codex and Work jumps to 500k tokens from 272k (about 80% more), and OpenAI won't charge long-context fees when that usage comes from a subscription. In the same hour Tibo said his Pro-plan poll surprised him: users chose to drop 5-hour limits over getting double weekly usage with them, so the hated 5-hour windows likely stay gone.

Tibo@thsottiaux

View on X ↗

Day 7/ We have increased default context length for GPT-6.1 Sol to 500k when used through Codex & Work, up from 272k or an ~ 80% increase. We will go higher in the future, but it helps reduce occurrence of compactions and with tasks that require keeping a lot of context and (reply) With this change, we are not charging long context fees for GPT-6.1 Sol when drawing usage from the subscription.

♥ 4,448⟲ 235💬 694👁 238.8k

Theo built himself a quota dashboard across his pooled accounts (11 credentials, including 6 Claude and 3 Codex) that charts remaining limits and forecasts when they'll run out, with per-account resets. Replies asked for it inside T3 Code, but given Theo's no-native-account-switching stance it looks like a personal tool for now.

Theo - t3.gg@theo

View on X ↗

Made a dashboard to track my Claude burn across accounts to (roughly) predict how long my limits will last me

♥ 176⟲ 2💬 30👁 8,105

v1.0.2 Stable moves Hubs out of beta (a coordinator plus 1–8 parallel workers across multi-repo projects, with a shared Library, queued work and human approvals) and adds dedicated chats, a live roster and run history for Claude and Codex subagents, plus per-chat provider accounts for Agent Gateway and external MCP.

Synara@trySynara

View on X ↗

Synara v1.0.2 Stable is out 🪽 🏡 Hubs hit Stable: 1–8 parallel workers, multi-repo projects, shared Library, queued work and human approvals. 🤖 Claude and Codex subagents: dedicated chats, live roster, run history, stop and background controls. ✅ Agent Gateway agents can manage Tasks to-dos right from chat. 🔌 Choose provider accounts for Agent Gateway and External MCP. 📚 Long chats reopen with recent history; load older messages on demand. 🧵 Clearer turn summaries, tool activity, Claude Monitor updates and background results. ⚡ Safer reconnects, recoverable drafts, and queued follow-ups that pause on Stop. 🔀 Fork from the exact message; save a different diff layout for each chat. 🧠 Claude resumes like Claude Code; OpenCode V2 no longer duplicates replies. 🛡️ Checkpoint Undo protects active work; better Windows launches and provider reliability. 📊 94.9% less initial history; 59/1,000 rail items; 65.5% fewer SQLite writes in synthetic tests.

♥ 20⟲ 1💬 5👁 1,087

T3 Code nightly now has a guided PR review mode, built by core dev @jullerino after a user asked for it. Julius says it's behind a feature flag while he quietly collects feedback on how it should work.

Faqih Muntashir@itsfaqih_

View on X ↗

Even though it's not using the https://t.co/Hh3B2eVm59, we have guided review in the nightly version now 🥳 Thank you so much! @jullerino

♥ 5⟲ 0💬 1👁 501

Subagents now get their own conversations: every Claude and Codex child agent appears in the right dock with its brief, current step, elapsed time and result, and you can open it, message it, background it or stop it.

Synara@trySynara

View on X ↗

Subagents now have their own conversations in Synara. Every Claude and Codex child agent shows up in the right dock with its brief, current step, elapsed time and result. Open it, message it, send it to the background or stop it.

♥ 1⟲ 0💬 0👁 33

Jay, CEO of OpenCode, described his ideal setup: an always-on Mac at home running a coding agent with computer use, session sync to his MacBook, and an iOS app with a voice agent paired to the Mac (1.2k likes, 117k views). Orca's CEO and official account replied that Orca already does all of it, and other replies pitched Codex Remote, cloud agents and Tailscale-plus-tmux setups.

Jay@jayair

View on X ↗

The setup I want: - Always-on Mac at home - Coding agent with computer use - MacBook that syncs my sessions - iOS app + voice agent paired with my Mac So I can voice control everything on my machine through the mobile app And work on those same session when I'm at my desk (reply from @JinjingLiang, Orca CEO) That's pretty much @orca_build

♥ 1,180⟲ 25💬 188👁 116.5k

Traycer v1.5.1: sign-ins and profiles now sync across hosts and devices, local-host lifecycle modes, Command Code as a new agent harness, continue any subagent as its own chat, chat auto-archive, a redesigned Activity sidebar with tab groups, CLI upgrades (live agent list, profile sync) and an experimental visual layout editor.

Traycer@TraycerAI

View on X ↗

Traycer v1.5.1 is live! 🚀 🖥️ Host lifecycle modes - Choose how your local host runs 🔄 Profile sync - Sign-ins and profiles move between hosts 🧩 Command Code - Now a supported agent harness 🗂️ Chat auto-archive - Chats archive themselves 💬 Subagent as chat - Continue any subagent in its own chat 📑 Activity sidebar - Sections, tab groups, Chrome-style dragging ⌨️ CLI upgrades - traycer profile, live agent list, worktrees 🧱 Visual layout editor - Experimental 📱 Faster on mobile - Quicker cold launch, lighter streams

♥ 7⟲ 1💬 1👁 282

Federico Viticci (MacStories founder, 85k followers) had Codex write custom firmware for a spare ESP32 gadget, creating "T3 Pebble", a tiny desk display that checks on his T3 Code agents in real time through T3 Code's orchestration API. It shows the API is open enough for outside hardware and tools to monitor running agents.

Federico Viticci@viticci

View on X ↗

had a spare ESP32 device so I had Codex create a custom firmware for it I now have this little guy that checks on @t3dotcodes in real-time via its orchestration API it's called T3 Pebble

♥ 29⟲ 0💬 7👁 2,959

Theo says native computer-use (an agent controlling a machine's screen directly) is coming to T3 Code, after Codex inside T3 Code picked a clumsy workaround to drive a remote machine.

Theo - t3.gg@theo

View on X ↗

I let codex (in t3 code) choose how to control the remote machine and this is what it chose. Needless to say, I am suffering. Can't wait til we get native CUA into T3 Code

♥ 11⟲ 0💬 2👁 653

Following his queue-vs-steer take, Theo showed that T3 Code has a setting to choose whether a new message sent mid-run queues up or steers the running agent by default.

Theo - t3.gg@theo

View on X ↗

btw you can change the default in T3 Code :)

♥ 12⟲ 0💬 1👁 2,261

v1.0.2 Beta 1: a full view of subagent runs (jump between parent and child chats, stop or background a run), forking from any message, and approving Hub worker requests from the coordinator chat, plus big speedups for long chats.

Synara@trySynara

View on X ↗

Synara v1.0.2 Beta 1 is out 🪽 🤖 Subagents, finally in full view: track active and finished runs, see results, jump between parent and child chats, or stop a run and send it to the background. 🧵 No more mystery turns: see clearer tool summaries, run times, background resumes and Claude Monitor updates. 📚 Pick up long chats where you left off, then page back through history — a seeded 2,005-message test cut initial thread-detail data by 94.9%. ⚡ Fewer lost or duplicated moments: streaming, reconnects, drafts and queued sends recover more reliably; Stop and Resume behave predictably. 🔀 Fork from the exact message you choose, without bringing later conversation context along. 🧭 Handle Hub worker approvals right from the coordinator chat — you stay in control of every approval. … [changelog trimmed here — full text on X]

♥ 4⟲ 2💬 1👁 166

Delta (Zed's agent workspace) now has @mentions: tag a teammate in a thread or comment and they get an OS notification plus an Inbox entry. The demo shows opening a mention, asking the agent for context in a subthread, then replying in the original thread, pushing Delta further toward team collaboration on agent threads.

Zed@zeddotdev

View on X ↗

Delta now has @mentions. Tag a teammate in a thread or comment and they’ll get an OS notification, with the mention saved in their Inbox. In this clip, Joseph opens a mention, uses a subthread to ask the agent for context, then returns to the original thread to comment.

♥ 7⟲ 0💬 0👁 648

T3 Code now supports GitCafe (a YC S26 git host) as a source-control provider, posted by core dev Julius. The settings screen shows a Source Control Providers list with GitHub, GitLab, Azure DevOps, Bitbucket, Forgejo/Gitea and GitCafe, each powered by that host's CLI.

Julius@jullerino

View on X ↗

You can now use @gitcafe in @t3dotcodes

♥ 38⟲ 3💬 3👁 994

Grok Build now has a /density command that sets how much of each tool call the transcript shows, from full diffs down to grouped one-line summaries. Spotted by feature-tracker account @techdevnotes.

Tech Dev Notes@techdevnotes

View on X ↗

Grok Build now has /density It controls how much of each tool call the transcript shows (detailed diffs through fully grouped one-line summaries)

♥ 8⟲ 0💬 2👁 950

Superset Mobile is now out on iPad, confirmed by CEO Kiet Ho with a tablet version of the mobile demo video. His Oct 9 'sent from my iPad' teaser of it drew about 20.7k views; Android is next on the roadmap after iPad and Cloud.

Kiet@FlyaKiet

View on X ↗

To celebrate our iPad launch, I edited our mobile video to make it tablet-friendly Like it or not, this is peak AI-native team performance

♥ 6⟲ 1💬 2👁 171

Orca v1.4.224 lets each automation run pick the agent's model and effort level, set in the automation editor under Advanced, Extra agent arguments, or with --extra-agent-args in the CLI (for example --model haiku --effort high).

Orca ADE@orca_build

View on X ↗

In the latest Orca v1.4.224, automations can now set the agent’s model and effort per run. In the automation editor, open Advanced → Extra agent arguments CLI: use --extra-agent-args with orca automations create/edit. Example: --model haiku --effort high

♥ 7⟲ 1💬 1👁 472

Contributor work in progress: sandboxes in T3 Code. Instead of running on main or a worktree, it spins up a fresh cloud VM, runs the agent there, and adds it as a remote environment on all your devices.

Christian Bager Bach Houmann@chrisbbh

View on X ↗

Worked on getting sandboxes into T3 Code Instead of running on main or a new worktree, it spins up a fresh https://t.co/Gdraws9Sza VM, run the agent there, and adds it as a remote environment on all devices No more managing memory on my machines or balancing worktrees https://t.co/p99ZDjBPmT

♥ 63⟲ 0💬 16👁 4,691

Devin now accepts personal ChatGPT plans (Go, Plus or Pro), so GPT usage inside Devin draws from your ChatGPT quota. Another big product joins the bring-your-own-subscription wave; OpenAI's Tibo framed it as 'your ChatGPT subscription is now also a Devin subscription'.

Cognition@cognition

View on X ↗

You can use any personal ChatGPT plan with Devin Just connect your Go, Plus, or Pro plan to Devin and all GPT model usage will draw down from you plan's quota https://t.co/7vVFQq3IAB

♥ 833⟲ 54💬 108👁 527.2k

T3 Code core dev Julius shows 'T3 Code Preview' appearing in macOS's default web browser list, hinting that T3 Code's built-in preview browser is becoming a full system browser target for links.

Julius@jullerino

View on X ↗

idk how to feel about this... https://t.co/cyIRDMwKL9

♥ 208⟲ 0💬 6👁 12.6k

Codex on Windows gets a new sandbox mode built on Microsoft's Execution Containers (MXC): faster setup, stronger network enforcement and finer-grained file access controls. Needs a compatible Windows 11 device.

OpenAI Developers@OpenAIDevs

View on X ↗

An update for builders using Codex on Windows: We’ve built a new sandbox mode using @Microsoft’s Execution Containers (MXC) for faster setup, stronger network enforcement, and granular file access controls. Requires a compatible Windows 11 device.

♥ 22⟲ 3💬 3👁 2,801

v1.0.1 stable: OpenCode v2, multiple project folders, Tasks in Stable, and cross-chat search.

Synara@trySynara

View on X ↗

Synara v1.0.1 Stable 🪽 🔌 OpenCode v2 support, with v1 servers still working. 📁 Work across multiple project folders with Codex or Claude in Local mode. ✅ Tasks comes to Stable: organize work, plan in Inbox and hand tasks to agents. Hubs remain Beta-only. 🔎 Search messages across chats and find actions or settings from workspace search. ⌨️ Archive, snooze or mark chats unread with shortcuts; preview drafts and undo “done.” 💬 Keep questions and answers in the transcript; recover interrupted work with clearer status and error details. ⚡ Smoother streaming, fewer idle chat updates and less contention between sessions. 🧠 Choose provider accounts for usage rings; disabled providers stay out of pickers. 🔋 Keep your Mac awake while agents work—or all the time. 🗂️ File tabs remember scroll position; project actions open in right-side terminals. 🎨 Synara is now the default theme, with clearer browser overlays and sharper sidebar text. 🛠️ Clearer Git and Kanban actions, better Markdown previews, and more reliable provider sessions.

♥ 32⟲ 2💬 8👁 1,868

Ultrafast for GPT-6.1 Sol is rolling out in the API, Codex and ChatGPT Work, at up to 8x the speed of Sol Standard.

OpenAI Developers@OpenAIDevs

View on X ↗

Ultrafast is rolling out today for GPT-6.1 Sol in the API, Codex, and ChatGPT Work. Near-Astra intelligence at up to 8x faster speeds than Sol Standard, so you can build as fast as the ideas come.

♥ 4,674⟲ 313💬 387👁 1.2M

Rewritten terminals: 13 ms typing lag (vs 24–40 ms in other ADEs they tested) and 5x faster big output. Rolling out in v0.13.5.

prateek - orchestrator.inc@agent_wrapper

View on X ↗

We are GUI and orchestration first, but honestly, terminals in @useOrchestrator weren't great. So I did what any developer would I pulled an all-nighter to fix that, and now they're faster than any ADE we tested: - Typing lag: 13 ms (Other ADEs: 24–40 ms) - Big output hits the screen 5× faster, and output-heavy - commands finish up to 14× sooner - Closing a terminal: ~40 ms, down from ~100 ms - Fixed the stray %, ghost tabs, flicker and tab reordering You can use these terminals to run your favorite editors, tools, or even more terminal agents! These an many more performance optimizations being rolled out in v0.13.5 Enjoy! PS: The video is NOT SPED UP

♥ 110⟲ 13💬 12👁 8,157

Every paid personal ChatGPT plan can now be used in Conductor through Sign in with ChatGPT.

Conductor@conductor_build

View on X ↗

thanks to Sign in with ChatGPT, now every paid personal ChatGPT plan can be used in Conductor!

♥ 41⟲ 0💬 2👁 6,864

Sign in with ChatGPT usage now covers the Go plan as well as Plus and Pro, across launch-partner and open-source apps.

Vignesh@vigyso

View on X ↗

BIG NEWS - Sign in with ChatGPT plan usage is now also available to all customers on the ChatGPT Go plan. That is - Customers on Go, Plus and Pro plans can now login to all commercial launch partner apps + open source applications and take their ChatGPT plan usage with them. So much more goodness is yet to come!

♥ 199⟲ 10💬 25👁 41.8k

Cross-environment orchestration is coming soon.

Julius@jullerino

View on X ↗

Cross env orchestration, soon in a T3 Code near you!

♥ 902⟲ 7💬 78👁 22.0k

Working on multi-PR and multi-issue support for Orca workspaces.

jinjingliang@JinjingLiang

View on X ↗

Working on some multi-PR / issues support for @orcabuild workspaces

♥ 20⟲ 1💬 3👁 1,842

Cline Cloud Agents run tasks in a cloud sandbox from your browser and open PRs.

Cline@cline

View on X ↗

1/ Introducing Cline Cloud Agents. You can now use Cline straight from your browser. Give it a task and it runs in a secure cloud sandbox, writing code, testing its work, then opening a PR on your GitHub repo.

♥ 204⟲ 16💬 12👁 12.1k

Claude Haiku 5.5 added. Cursor says it costs 10x less than Haiku 4.5 on shorter requests.

Cursor@cursor_ai

View on X ↗

Claude Haiku 5.5 is now available in Cursor! On shorter requests, it costs 10x less than Claude Haiku 4.5. Turn it on in Cursor Settings > Models.

♥ 2,690⟲ 125💬 157👁 240.9k

Claude Haiku 5.5 launched, at about 75% lower average cost than Haiku 4.5.

Claude@claudeai

View on X ↗

Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we’ve ever released. On average, it costs around 75% less to run than Claude Haiku 4.5.

♥ 44.3k⟲ 3,202💬 1,530👁 6.1M

The T3 Code MCP works with any connector, so you can drive threads from ChatGPT, Claude or Grok.

Julius@jullerino

View on X ↗

The T3 Code MCP now works with any connector. Add it to ChatGPT, Claude, Grok, etc. and drive your threads from there.

♥ 1,440⟲ 44💬 87👁 48.9k

Agent Search is now in the CLI (`orca search`), so agents can search past Claude Code, OpenCode and Grok sessions using BM25 ranking.

Orca ADE@orca_build

View on X ↗

Agent Search 2.0 is also available through the Orca CLI: `orca search <your keywords>` Your agent can use it too! Search past conversations across Claude Code, OpenCode, Grok, and more in one place. Indexed search makes finding specific conversations fast, and our built-in BM25 algorithm ranks the most relevant matches first.

♥ 78⟲ 2💬 9👁 5,686

Paperclip v2026.1005.0 (Oct 5): Grok Build on its native runner, agents keep their own files (AGENTS.md) across tasks, company skills sync from GitHub repos, and Browser Use Cloud gives agents a live browser people can watch and take over inside the task. Also one-click connectors, two-way Slack threads, and a per-message model/effort picker.

Paperclip@papercliping

View on X ↗

Paperclip v2026.1005.0 is out 🤖 Grok Build on the native runner 📁 Agents keep their own files across tasks 🧩 Skills synced from GitHub repos 🌐 Browser Use Cloud, live in the task 179 commits from 7 contributors https://github.com/paperclipai/paperclip/releases

♥ 41⟲ 3💬 2👁 2,216

Webhook-triggered tasks, with an optional T3 Connect relay that holds and replays events for up to 24h.

Julius@jullerino

View on X ↗

You can now create webhook triggered tasks in @t3dotcodes! Perfect for issue triaging or other tasks you wanna run on a trigger. Signed in to T3 Connect? You get a URL that forwards events to your environment, or opt in to have them held and replayed for up to 24h for added durability if it's down. No Connect? Bring your own public URL. For now it's quite "raw" (checks hmac signatures and forwards the event to your prompt template), but will add some UX sugar on top in coming releases. If you don't use T3 Connect you can of course also bring your own public URL that triggers the task.

♥ 784⟲ 19💬 51👁 28.0k

Deleted workspaces can be restored from the workspaces tab, bringing back files, chats and the PR.

Kiet@FlyaKiet

View on X ↗

Accidentally delete a workspace? You can now restore them in the workspaces tab. Your files, chats, and PR will all come back. try it now in latest version of @superset_sh

♥ 17⟲ 1💬 2👁 787

Remote Server 2.0 is coming this week.

Orca ADE@orca_build

View on X ↗

Remote Server 2.0 takes the stage this week 👀

♥ 127⟲ 5💬 10👁 9,122

Testing an Artifacts tab as the default right panel, where agents package a visual summary of their work instead of you reading diffs.

Charlie Holtz@charlieholtz

View on X ↗

experimenting with an Artifacts tab as the default right panel in every workspace. the agents are good enough now that you don't need to look closely at the diffs. instead, the agent can package up a quick + beautiful overview its work as an artifact

♥ 199⟲ 3💬 17👁 8,959

v1.0.0 stable: Inbox, PR/issue review, multiple accounts, switching providers inside a conversation, and Kanban.

Synara@trySynara

View on X ↗

Synara v1.0.0 is here 🪽 The biggest update ever made 🤖 Oh My Pi is now available in Stable, with runtime-discovered models and thinking levels. 📥 Inbox brings active work, chats needing attention, review requests and a daily activity recap together; Auto-fix CI can tackle failing checks on your PRs. 🔌 Configure multiple accounts per provider, sign in from Settings, see per-account usage and choose models from separate account tabs. 🔀 Switch providers inside the same conversation while keeping its chat and workspace context. … [changelog trimmed here — full text on X] 🧪 Stable keeps Kanban; Hubs and the Tasks to-do list remain Beta-only, with separate apps, data and update feeds.

♥ 65⟲ 8💬 4👁 6,458

Delta v0.18 adds thread bookmarks.

Zed@zeddotdev

View on X ↗

As of Delta v0.18, you can bookmark threads with `cmd-d` on macOS and `ctrl-d` on Windows and Linux. Filter your threads to show only bookmarked ones using the Show Bookmarked Threads button in the sidebar.

♥ 114⟲ 1💬 1👁 7,181

Agent Search 2.0: search full session history, including tool calls, and press Enter to jump back into a session.

Orca ADE@orca_build

View on X ↗

Introducing Orca Agent Search 2.0 ⚡ That agent conversation you can almost remember? Find it in an instant. Search the full session history, including tool calls, then hit Enter to jump straight back in.

♥ 252⟲ 11💬 16👁 16.4k

SDK: you can replace Cursor's system prompt with your own (rolling out account by account).

Cursor@cursor_ai

View on X ↗

You can also replace Cursor's system prompt with your own. Rules, skills, and tool schemas still load. We're enabling it account by account.

♥ 37⟲ 1💬 3👁 9,635

SDK: custom tools can carry MCP annotations such as readOnlyHint and destructiveHint.

Cursor@cursor_ai

View on X ↗

Custom tools can now carry MCP annotations like readOnlyHint and destructiveHint, so the model can tell a lookup from a delete before it calls one.

♥ 44⟲ 0💬 5👁 4,463

SDK: background subagents report results back to the parent run.

Cursor@cursor_ai

View on X ↗

Background subagents now report back. Their results return to the parent as a follow-up turn on the same run, so stream() and wait() carry through until every subagent finishes.

♥ 89⟲ 2💬 8👁 10.8k

SDK: run.steer() lets you steer agents mid-run, and busy subagents move to the background and keep working.

Cursor@cursor_ai

View on X ↗

You can now steer Cursor SDK agents while they run. run.steer() adds your message to the next turn. If a subagent is mid-task, it moves to the background and keeps working.

♥ 1,335⟲ 84💬 142👁 132.2k

Reminder that Orca has native GitHub, GitLab, Linear and Jira integrations.

Orca ADE@orca_build

View on X ↗

Hey friends! How's the weekend going? ✨ Are you building or touching grass? Maybe touching grass AND agents building at home or a VPS? 😆 Did you know Orca has native integrations for GitHub, GitLab, Linear and Jira?! Using them? Running automations? Any feedback? 🙏

♥ 76⟲ 3💬 21👁 5,478

Port forwarding: an app running on a remote box opens at the same localhost port on your laptop.

Traycer@TraycerAI

View on X ↗

Once your agents run on another machine, "localhost" stops meaning anything. Introducing Port Forwarding in Traycer. The agent builds and runs the app on the big box, then hands you the port. Open localhost:port on your laptop. Same port. No tunnels. Nothing public.

♥ 9⟲ 2💬 0👁 696

v0.13.3 adds a dedicated orchestrator agent and kanban across harnesses and subscriptions, an in-app browser with annotated feedback, a diff sidebar and a mobile app.

prateek - orchestrator.inc@agent_wrapper

View on X ↗

for the last month, we have been shipping non-stop in silence @useOrchestrator v0.13.3 is out - manage agents across harnesses and subscriptions with a dedicated orchestrator agent and kanban view (claude projects style) - your agent shows you it's work in the in app browser (give it annotated feedback right inside the app) - explore changes / file diffs in the sidebar - take your agents with you using the mobile app (orchestrate on the fly) Check us out!

♥ 36⟲ 6💬 3👁 3,177

OV2 also adds inline subagents, a server-side queue that survives restarts, thread forking, auto-resume after limits reset, and scheduled tasks.

Julius@jullerino

View on X ↗

- Subagents show up inline with their model, status and results - Stop actually ends background work - Server-side queue: steer or queue, survives restarts - Fork any thread, natively where the provider supports it - Auto-resume when usage limits reset - Scheduled tasks

♥ 326⟲ 1💬 2👁 19.6k

Orchestrator V2, a rewrite of the orchestrator that runs your agents, is live in nightly.

Julius@jullerino

View on X ↗

T3 Code Orchestrator V2 is now live in nightly. We rewrote the orchestrator that runs your agents. Here's what that gets you 🧵

♥ 1,899⟲ 37💬 85👁 113.0k

Orca Mobile does speech-to-text through your paired desktop app, using local models or low-cost OpenAI cloud options.

Orca ADE@orca_build

View on X ↗

Typing on the go? Hard pass ⌨️ ❌ Orca Mobile turns speech into text via your paired desktop app. Just tap send. No mobile setup. 🎙️ ✅ With several local models. Also low-cost OpenAI cloud options. Perfect for languages where local models occasionally get confused 🫠

♥ 83⟲ 3💬 11👁 6,273

Delta handoff: catch up on a teammate's thread through a subthread.

Zed@zeddotdev

View on X ↗

Handoff in Delta is fun: 1. A teammate sends you a link to their thread. 2. You ask the agent to catch you up in a subthread, without rereading or polluting the main thread. 3. You get to work. That’s how Cole jumped into a massive thread started by Antonio and Nathan. Read about it here. delta.dev/blog/planning-together-in-delta

♥ 114⟲ 5💬 7👁 10.7k

Rollouts finds the PR behind a regression and opens an issue. One click starts a cloud agent to fix it.

Cursor@cursor_ai

View on X ↗

When Rollouts catches a regression, it finds the offending PR and opens an issue. One click starts a cloud agent to fix it. Rollouts usage credits are included through Oct 3.

♥ 860⟲ 44💬 93👁 74.9k

Paperclip v2026.1001.0: any agent can become a scheduled GitHub PR review bot, approvals and answers now queue while an agent is mid-run instead of bouncing, agents get persona avatars across the app, and a Railway connector adds deploy/rollback tools. Harnesses now default to full-auto.

Paperclip@papercliping

View on X ↗

Paperclip v2026.1001.0 is out 🛡️ Hardened native runner + chat recovery 🎭 Agent personas across the app 🔍 Agents as GitHub PR review bots 🚂 Railway connector + queued approvals 77 commits from 8 contributors

♥ 58⟲ 2💬 7👁 3,554

GLM 5.3 and GLM 5.3 Flash added. Cursor says GLM 5.3 Max is the top-scoring open-weight model on CursorBench 4.0.

Cursor@cursor_ai

View on X ↗

GLM 5.3 and GLM 5.3 Flash are now available in Cursor! GLM 5.3 Max is the best-scoring open-weight model on CursorBench 4.0.

♥ 3,726⟲ 181💬 210👁 254.4k

Conductor Mobile is live on the iPhone App Store for running a team of cloud agents from your phone.

Charlie Holtz@charlieholtz

View on X ↗

Introducing Conductor Mobile! Run a team of cloud agents from your iPhone, live now in the App Store.

♥ 757⟲ 26💬 63👁 69.5k

Superset Mobile is coming soon to iPad and iPhone Duo.

Kiet@FlyaKiet

View on X ↗

Superset Mobile coming soon to the iPad and Duo 📲 @superset_sh

♥ 27⟲ 1💬 3👁 2,310

Community build: NetworkChuck built an "AI IT department" in Paperclip, with Claude Code, Hermes, Codex and local models in assigned roles, sharing tasks and gated by human approvals (YouTube video).

Paperclip@papercliping

View on X ↗

@NetworkChuck built an AI IT department with Paperclip. Claude Code, Hermes, Codex, and local models working as a team—with assigned roles, tasks, and human approvals. Watch him put it together:

♥ 33⟲ 2💬 7👁 3,841

Deleted projects can be restored for up to 30 days. Before you delete, Superset shows which agents and terminals are still running.

Kiet@FlyaKiet

View on X ↗

Accidentally deleted a project in @superset_sh? You can restore it for up to 30 days. Hit Undo right after deleting, or head to Settings → Projects → Recently deleted to bring the project and its workspaces back. Before you delete, Superset also shows which agents and terminals are running, so you can see whose work you’d interrupt. Available in the next version of superset.sh

♥ 11⟲ 1💬 2👁 790

Account switcher and usage tracking: see Claude and Codex usage and rate-limit resets, and hot-swap Codex accounts without logging in again.

Orca ADE@orca_build

View on X ↗

Glad you ask, yesss! 👌 We've got an Account Switcher & Usage Tracking ✨ See Claude and Codex usage, rate-limit resets, and hot-swap Codex accounts without re-logging in.

♥ 21⟲ 0💬 2👁 2,547

Zed v1.22: the agent can spawn subagents on different models.

Zed@zeddotdev

View on X ↗

🚀 Zed v1.22 is out! Zed's agent can now spawn subagents with different models on the fly. Each subagent card shows which model it's using. Thanks itsfuad!

♥ 612⟲ 25💬 23👁 29.7k

Orchestration can be installed as a skill (Settings -> Orchestration -> Install Skill).

Orca ADE@orca_build

View on X ↗

Have you ever used the Orchestration feature? It's so simple and so powerful. Take a couple of minutes and try: Settings -> Orchestration -> Install Skill Pick a case, agent will know what to do 🚀 Please share your cases or questions 🤗

♥ 174⟲ 10💬 16👁 14.6k

Previewhost gives each agent worktree its own local preview environment with one URL, and also works over MCP.

rojhat@0xRojhat

View on X ↗

I released Previewhost this week. When multiple coding agents work in separate Git worktrees, running what they build locally becomes exhausting. Managing ports, starting services, setting up env variables and development secrets again... Previewhost gives each task its own local environment. It can start your frontend, backend, and databases, connect them together, and give you one human-readable local URL to test the result. It also works through MCP, so agents can start and inspect their own preview environments directly.

♥ 1⟲ 0💬 2👁 317

T3 Code is a launch partner for Sign in with ChatGPT, with multiple ChatGPT accounts supported.

T3 Code@t3dotcodes

View on X ↗

getting started with T3 Code has never been easier, excited to be a launch partner of Sign in with ChatGPT! install t3 code, sign in to your ChatGPT accounts (yes… plural, t3 code supports multiple accounts), and send your first prompt. no external dependencies required. available now on latest t3 code!

♥ 1,409⟲ 34💬 48👁 135.7k

Remote hosts: agents on different machines can message each other, forward ports and copy files.

Traycer@TraycerAI

View on X ↗

🖥️ Remote hosts Agents on different computers can now work together: • Port forwarding: an agent starts a dev server on a remote machine, and it opens on your laptop at localhost. • Agents can message agents on your other computers, create new ones there, run commands and copy files between them. • When a remote agent opens a browser, the tab appears on the computer you messaged it from, with that machine's localhost and sign-ins. • Sign in Claude Code, Codex or Grok on a remote host with a one-time code. See and stop port forwards in the chat's Background panel or in Settings › Host › Overview › Ports.

♥ 1⟲ 0💬 1👁 170

Model routing switches account or model, or waits, when you hit a usage limit.

Traycer@TraycerAI

View on X ↗

🔀 Model routing Hit a usage limit, got signed out, or the provider is down? Traycer automatically: • switches to another account for the same agent • moves to an equivalent model in the same tier (frontier, flagship or standard) • or waits for the limit to reset Every switch has a countdown you can cancel. Turn on Route automatically in Settings › Model routing.

♥ 2⟲ 0💬 1👁 224

v1.4.0: model routing, Auto mode, version history, remote hosts, themes, labels, Antigravity support and mobile apps.

Traycer@TraycerAI

View on X ↗

Traycer v1.4.0 is live! 🚀 🔀 Model routing - ran out of usage? Automatically switches your chat to another account or model. 🛡️ Auto mode - AI approves routine commands for you and asks before risky ones. 🕰️ Version history - every change to your specs and tickets is saved, so you can always go back. 🖥️ Remote hosts - agents on different computers can talk to each other and forward ports, so what runs on one opens on the other. 🎨 Themes & layout - pick a theme or import one from VS Code, set a wallpaper, and choose what your workspace shows. 🏷️ Labels - tag your tasks so they're easy to find. ✨ Antigravity - Google's coding agent now works in Traycer. ⚡ Faster - far less memory, far fewer freezes. 📱 Mobile - Android on Google Play, iPhone on TestFlight. And a lot more! Changelog: github.com/traycerai/traycer/releases/tag/desktop-v1.4.0

♥ 23⟲ 5💬 7👁 1,434

The Codex CLI also adds /usage analytics, managed worktrees, Mermaid/LaTeX rendering and collapsible diffs.

OpenAI Developers@OpenAIDevs

View on X ↗

There’s a lot more for builders packed into the Codex CLI update: – /usage for activity and usage analytics – New themes to choose from with /theme – Support for managed worktrees – Mermaid diagrams and LaTeX rendered directly in the terminal – Collapsible diffs and tool output – Dedicated areas for status, hints, and keyboard shortcuts

♥ 101⟲ 4💬 1👁 15.2k

Major Codex CLI refresh: full-screen interface and ways to manage parallel work.

OpenAI Developers@OpenAIDevs

View on X ↗

Codex CLI just got a major refresh, with a new look and powerful new capabilities. We’re continuing to invest in Codex CLI for developers who live in the command line. New full-screen interface, better readability, and ways to manage parallel work, all inside your terminal.

♥ 4,140⟲ 184💬 111👁 259.7k

Codex cloud environments: reusable environments, so agents keep working after you close your laptop.

OpenAI Developers@OpenAIDevs

View on X ↗

You can finally close your laptop now and your agents will keep working. Codex cloud environments are here. Reusable environments mean less setup and faster starts, with your repo, dependencies, scripts, and settings already in place.

♥ 5,653⟲ 383💬 267👁 660.5k

Conductor partnered with OpenAI on Sign in with ChatGPT, so you can bring your Codex subscription in a few clicks.

Charlie Holtz@charlieholtz

View on X ↗

We've partnered with OpenAI to bring Sign in with ChatGPT to Conductor! You can now bring your Codex subscription to Conductor in a few clicks:

♥ 243⟲ 10💬 24👁 17.4k

Sonnet 5.5 is live in Conductor.

Charlie Holtz@charlieholtz

View on X ↗

Sonnet 5.5 is live in @conductor_build! I have a feeling Opus 5.5 will still be my default but this chart is certainly surprising...

♥ 49⟲ 2💬 7👁 3,914

Sonnet 5.5 is available in Cursor, and Cursor says it performs on par with Opus on many tasks.

Cursor@cursor_ai

View on X ↗

Sonnet 5.5 is now available in Cursor! It’s a strong model, performing on par with Opus in many tasks.

♥ 3,156⟲ 126💬 147👁 226.6k

Flags when an open file changes on disk because an agent edited it, and shows both versions side by side. Also adds autosave plus drag and rename in the file tree.

Kiet@FlyaKiet

View on X ↗

Stop losing edits when your agents touches the same file. @superset_sh now flags when an open file changes on disk and shows both versions side by side. Keep editing, reload, or overwrite. Plus autosave, drag and rename in the file tree, thanks to @theblondealex

♥ 14⟲ 2💬 3👁 1,008

Worktrees are no longer required. New "local workspaces" point straight at the project root while still organizing agent sessions.

Kiet@FlyaKiet

View on X ↗

We're no longer requiring worktrees in @superset_sh While worktrees are powerful, we found many users prefer to work in the root directory of their project, using workspaces to organize their agent sessions. Hence, you can now create "local workspaces" which points to the root of your project. Just select local for the workspace type.

♥ 86⟲ 3💬 13👁 8,890

Orca can index your agents' full chat history for local search (Settings > Agent Session Search).

Orca ADE@orca_build

View on X ↗

Yes! Orca can now index your agent's full chat history, making it easy to search any past conversation. Search indexes are stored locally on your machine. You can enable this feature under "Settings > Agent Session Search"

♥ 183⟲ 8💬 9👁 15.0k

Claude Code cross-session messaging: one session can list your other Claude Code sessions (desktop, terminal, even other machines or cloud via Remote Control) and send them a message, like "tell the payments session the schema changed". It shows up as a card labeled with the sender, waits if the target is mid-task, and the target can reply. Shipped around early September; resurfaced with this week's desktop session-manager release (docs: code.claude.com/docs/en/cross-session-messaging).

Arman Obosyan@armanobosyan

View on X ↗

Claude Code cross-session messaging is insane. It’s crazy to watch two independent sessions discuss and coordinate optimizations for Sugra-OS and Azure-Opt in real time. They’re not just running in parallel anymore - they’re actually arguing, challenging each other, and https://t.co/LFQKB1hNAZ

♥ 4⟲ 0💬 0👁 411