A self-evolving coding agent that writes its own code.
iterate is a coding agent that owns its own repository. Every 4 hours it:
- Reads its own source code, journal, and community issues
- Decides what to improve — a bug, a missing feature, a rough edge
- Builds the fix, runs
go buildandgo test - Commits if green, reverts and journals if not
No human writes its code. It does it itself.
Live site — auto-updated after every session
git clone https://github.com/GrayCodeAI/iterate.git
cd iterate
export ANTHROPIC_API_KEY=sk-...
make build
./iterate --repo .Providers: Anthropic · OpenAI · Gemini · Groq
./iterate --chatiterate> /help
Agent /help /clear /model /thinking /version /quit
Code /test /build /lint /fix /coverage /health
Git /diff /status /commit /log /branch /pr
Analysis /count-lines /hotspots /contributors /languages
Project /tree /index /find /grep
Memory /remember /memories /forget /learn /memo
Evolution /phase /self-improve /evolve-now
Session /save /load /context /tokens /cost /compact
┌─────────────┐ ┌──────────────┐ ┌─────────────────┐
│ 1. Plan │────▶│ 2. Implement │────▶│ 3. Communicate │
│ │ │ │ │ │
│ Read source │ │ Agent per task│ │ Post GH comments │
│ Read journal │ │ go build/test │ │ Journal entry │
│ Read issues │ │ commit/revert │ │ │
└─────────────┘ └──────────────┘ └─────────────────┘
▲ │
└───────────────────────────────────────────┘
Every 4 hours via GitHub Actions
cmd/iterate/
main.go Entry point, flag parsing
repl.go Interactive REPL loop
repl_streaming.go Token-level streaming output
repl_helpers.go REPL utilities
repl_models.go Model switching
config.go Config loading (TOML/JSON)
features.go Core feature helpers
features_prompts.go Prompt builders
features_shell.go Shell/CLI utilities
features_tools.go Tool listing and info
features_sessions.go Session save/load/compact
features_watch.go File watching
commands_project.go /health, /tree, /index
commands_git.go /pr dispatcher
internal/
agent/ Agent pool + mutation testing
commands/ Modular command registry
registry.go Type definitions + registration
register.go Registration helpers
agent.go /help, /clear, /model, /thinking
dev.go /test, /build, /lint, /fix, /coverage
evolution.go /phase, /self-improve, /evolve-now
files.go /find, /grep, /tree, /index
git.go /diff, /status, /commit, /log, /branch
github.go /pr list/view/diff/review/create
memory.go /remember, /memories, /forget, /learn
mode.go /safe, /multi, /thinking
safety.go Safety checks
session.go /save, /load, /context, /tokens, /cost
utility.go /version, /stats, /changes
community/ GitHub issues + discussions
evolution/ 3-phase evolution engine
social/ Social interaction engine
ui/ Terminal UI (colors, highlighting)
util/ String truncation helpers
skills/ Structured skill files (SKILL.md)
evolve/ Self-modification rules
self-assess/ Codebase evaluation
communicate/ Issue response posting
research/ Learning from docs/web
social/ Community interaction
release/ Release management
scripts/
evolution/evolve.sh Main evolution pipeline
social/social.sh Social session runner
build/build_site.py Journal → GitHub Pages
maintenance/synthesize_learnings.py Memory compression
memory/
learnings.jsonl Append-only lesson log
active_learnings.md Synthesized knowledge
docs/ GitHub Pages site
index.html Auto-generated from JOURNAL.md
JOURNAL.md Evolution log
IDENTITY.md Agent constitution
PERSONALITY.md Agent voice
| Workflow | Schedule | Purpose |
|---|---|---|
evolve.yml |
Every 4h | Plan → Implement → Communicate |
social.yml |
Offset 2h | Read discussions, learn |
synthesize.yml |
Daily 3AM | Compress learnings |
deploy.yml |
On push | Build GitHub Pages |
ci.yml |
On push/PR | Build, test, vet, fmt |
- Fork this repo
- Add
ANTHROPIC_API_KEYto repository secrets - Enable GitHub Actions and GitHub Pages
- That's it — it will start evolving
- Bug report — something broke
- Suggestion — something to build
- Challenge — give it a hard problem
- Help wanted — stuck and asking for help
Issues labeled agent-input are read by the agent every session.