A team that never trips over itself.
Every worker gets its own copy of your repo. Only what passes review comes back.
How Flock works
Meet the flock.
The IDE where a team of AI agents writes, tests, debugs and ships your code.
You watch every move.
One real run, replayed
Fifteen agents. Fifty-five seconds.
Every bar is a real agent, every second is real. Drag back to replay it.
- elapsed
- 0:55
- agents ran at once
- 15
- tests passing
- 15 / 15
- the whole job
- $0.72
- Lead : planned for 16 seconds, waited for the workers, then checked their work until the end.
- Caesar : 5.1 s ; tests pass.
- Search : 4.4 s ; tests pass.
- Luhn : 5.3 s ; tests pass.
- Brackets : 5.6 s ; tests pass.
- Escape : 5.7 s ; tests pass.
- Anagram : 5.4 s ; tests pass.
- Rle : 6.1 s ; tests pass.
- Wordfreq : 5.9 s ; tests pass.
- Matrix : 5.7 s ; tests pass.
- Levenshtein : 6.7 s ; tests pass.
- Roman : 7.1 s ; tests pass.
- Primes : 7.1 s ; tests pass.
- Lru : 8.3 s ; tests pass.
- Base64 : 8.9 s ; tests pass.
- Semver : 13.1 s , $0.069, its tests pass, first try; tests pass.
0:55
[{"name":"Caesar","seconds":5.1,"end":22.46,"detail":null},{"name":"Search","seconds":4.4,"end":22.73,"detail":null},{"name":"Luhn","seconds":5.3,"end":23.25,"detail":null},{"name":"Brackets","seconds":5.6,"end":23.52,"detail":null},{"name":"Escape","seconds":5.7,"end":23.71,"detail":null},{"name":"Anagram","seconds":5.4,"end":23.72,"detail":null},{"name":"Rle","seconds":6.1,"end":24.11,"detail":null},{"name":"Wordfreq","seconds":5.9,"end":24.48,"detail":null},{"name":"Matrix","seconds":5.7,"end":24.62,"detail":null},{"name":"Levenshtein","seconds":6.7,"end":24.94,"detail":null},{"name":"Roman","seconds":7.1,"end":25.35,"detail":null},{"name":"Primes","seconds":7.1,"end":25.37,"detail":null},{"name":"Lru","seconds":8.3,"end":27.11,"detail":null},{"name":"Base64","seconds":8.9,"end":27.19,"detail":null},{"name":"Semver","seconds":13.1,"end":30.95,"detail":{"seconds":13.1,"cost":0.069,"command":"cargo test --test semver","note":"its tests pass, first try"}}] - planning
- reading
- writing
- testing
- checking
Claude Opus · 15 modules · 15/15 tests · run 2026-11-11 · How we measured →
Features
It doesn't say it works. It shows you.
Agents open your app in a real browser, click through it and read the console, right next to your code.
Watch it test
Bugs, caught where they happen.
The agent sets the breakpoint, reads the stack and finds the null, in your debugger, where you can take over.
See the debugger
Every change is yours to keep.
Each edit lands as a full diff. Keep a hunk, comment on a line, send it back to the agent that wrote it.
Review changes
And so much more.
Your agents, over every app.
A small dock floats above whatever you're doing. Answer, steer or stop them without switching windows.
frames drawn
when nothing moves.
Written in Rust, drawn on your GPU. Not a web page.
Just say it.
Hold Ctrl+Space and talk. It's transcribed on your own machine.
Right where you left off.
Every chat in every project comes back, even after a restart.
- Claude
- ChatGPT
- Gemini
- OpenRouter
- Ollama
Bring the plan you have.
Sign in with Claude, ChatGPT or Gemini, or any API key.
Also: MCP servers you've set up elsewhere are found automatically. Images open in place. Files, Git, terminal, debugger and tests, all built in.
