Free and Open Source Nerve Center for Agentic Coding https://teapaw.org/t.co/UeKIMh52Yf
San Francisco
Joined April 2024
- Tweets661
- Following9
- Followers5.9K
- Likes790
Pinned Tweet
Today, we're launching the Traycer Desktop App.
Free. Open source. Built with AI orchestration in mind.
What’s new:
• Bring your existing subscription (Claude, Codex, Opencode, etc.)
• Agent-to-Agent communication and Loops
• Persistent workspaces with tabs and sub-tabs
• Share tasks and collaborate with teammates
Capability stopped being the hard part.
The environment is the new frontier, and that’s where Traycer is.
Traycer retweeted
people asked: "what's your factory's revert rate?"
0.
453 PRs merged by our factory. not one reverted.
so we dug deeper. how many needed any follow-up fix at all?
> 420 shipped clean
> 24 tripped a post-merge CI check. it fixed 22 itself
> 5 fixed only part of a problem. a follow-up got the rest
> 4 real bugs
3 boring rules:
1/ receipts or it doesn't merge. bug before, fix after, + what it couldn't check.
2/ agents don't grade their own work. fresh reviewers never see the author's conclusion, and on tricky PRs they debate each other to cover every angle.
3/ the factory ships the boring stuff. anything the agents flag as high-risk goes to a human.
week one, 12% of merges broke a check, since then, 3%.
it messes up, fixes it, messes up less.
pretty proud of this one ❤️
No single model is best at everything. We built this so you can bring the agents you already pay for into one workspace, just like @bemeroy says. They run side by side, ask each other questions and hand off work across chats.
switching models mid-task like @ImmonenPietari does here is why we built the shared workspace. claude and codex share the same filesystem and task history, so you never have to rebuild context when you hand off work.
Free to download.
App Store: apps.apple.com/us/app/trayce…
Play Store: play.google.com/store/apps/d…
Why choose when you can have them all?
Traycer is the best place to get the most out of each one.
A couple months ago I only had Claude and Codex subscriptions, but now that I discovered @TraycerAI I now have Claude, Codex, MiniMax, Xiaomi and Kimi - but hey that's the price you got to pay to have like 40+ agents running at across 5 projects at once. 😂
We are just getting started 🫡
Traycer retweeted
11 days ago we turned on our self evolving software factory.
since then it has merged 332 PRs it opened on its own. half of everything we shipped.
user bugs, sentry errors, red builds, flaky tests: each one wakes an agent, it triages the issue and the agent shows up with a tested PR.
43 of those PRs were the factory fixing and improving itself. That’s our KPI.
Coding agents just got free.
Flint from @Recursive_SI is live in Traycer via @OpenRouter.
Introducing Flint, @Recursive_SI’s first model: a fast local coding model built with our automated research system, the Eureka Machine.
• Best-in-class fast local coding
• Up to 2x faster than comparable models: 125+ tokens/sec on a MacBook Pro, 150+ on RTX Spark
• Built from scratch in under 4 months
Private, unmetered intelligence on your own machine. Open weights available now on @huggingface, run it locally on your favorite framework or try it for free on @opencode or @pidotdev.
stop hunting through menus. ⌘K.
type a few letters to run any command, jump to a task or open a settings section. actions without a shortcut live there too, and each command shows its shortcut so you pick them up as you go.
docs.traycer.ai/settings/key…
don't let a full context window sneak up on you.
a chip next to the composer shows how much context is left. when the agent supports it, one click compacts the conversation and frees up space.
docs.traycer.ai/panels/agent…
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.
docs.traycer.ai/panels/agent…
S tier 🫡
next stop, god tier.
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.
docs.traycer.ai/concepts/tas…
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.
docs.traycer.ai/panels/comme…
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.
docs.traycer.ai/panels
prefer the keyboard? ⌘D splits the group, then pick what goes in the new pane, like a terminal next to the chat.
docs.traycer.ai/settings/key…
your laptop doesn't have to be the machine running the agents.
add a host on another computer with three commands, approve the sign-in, and make it your active host. agents then run and work with files over there.
docs.traycer.ai/concepts/hos…