An autonomous AI coding agent that lives in your terminal.
No subscriptions. No cloud. No Python. Just npm install -g sharrowkin and go.
sk "add Redis caching to all GET endpoints"
Sharrowkin reads your codebase, plans a solution, writes the code, runs tests — and tells you exactly what it did and why. All on your machine.
› add Redis caching to all GET endpoints
Reading file…
└ database.py
Listing files…
└ src/routes/
Writing file…
└ src/cache.py +52 lines
Editing file…
└ 11 route files
Running command…
└ $ pytest -x -q 47 passed
● Added @cache decorator to 11 GET endpoints.
Skipped auth routes. TTL: 60s products · 30s users.
✦ 14.2s
| Cursor | Claude Code | Sharrowkin | |
|---|---|---|---|
| Runs locally | ✗ | ✓ | ✓ |
| Open source | ✗ | ✗ | ✓ |
| Free models (Ollama, freemodel.dev) | ✗ | ✗ | ✓ |
| Zero telemetry | ✗ | ✗ | ✓ |
| Pure TypeScript, no Python | — | ✗ | ✓ |
| Price | $20/mo | $20/mo | Free |
npm install -g sharrowkinOr run without installing:
npx sharrowkin "refactor auth module to use JWT"# Set your API key (freemodel.dev is free)
sk config set baseUrl https://cc.freemodel.dev
sk config set apiKey your_key_here
# Run a task
sk "add input validation to all API endpoints"
# Interactive mode
sk
# Switch model with arrow keys
/modelFree (no credit card):
kr/claude-sonnet-4-7 via freemodel.dev
kr/claude-opus-4-8 via freemodel.dev
llama3, qwen3 via Ollama (local)
Paid:
claude-opus-4-8 via Anthropic
gpt-5.5 via OpenAI
gemini-2.5-pro via Google
any OpenRouter model
Pure TypeScript. No backend. No server. The agent runs in the same process as the CLI.
sharrowkin-cli/
├── src/
│ ├── agent/
│ │ ├── agent.ts ReAct loop — think, act, observe
│ │ ├── tools.ts read · write · run · search
│ │ └── llm.ts fetch → any OpenAI-compatible API
│ ├── commands/
│ │ ├── run.ts task execution + terminal UI
│ │ └── watch.ts live observer mode
│ ├── ui/
│ │ └── welcome.ts the beautiful box
│ └── core/
│ ├── config.ts settings
│ └── types.ts AgentEvent types
└── frontend/ web UI — coming soon
Agent loop — standard ReAct:
- Think — LLM plans next action
- Act — executes one tool (read / write / run / search)
- Observe — feeds result back to LLM
- Repeat until
attempt_completion
| Tool | What it does |
|---|---|
read_file |
Read any file, with line numbers |
write_file |
Create or overwrite a file |
str_replace |
Surgical in-place edit |
list_files |
Directory listing |
run_command |
Shell with timeout |
search_files |
Grep across codebase |
attempt_completion |
Signal task done |
sk config # view all settings
sk config set model kr/claude-opus-4-8
sk config set baseUrl https://cc.freemodel.dev
sk config set apiKey your_keyOr edit ~/.config/sharrowkin-cli/config.json directly.
git clone https://github.com/narelabs/sharrowkin-cli
cd sharrowkin-cli
npm install
npm startPRs welcome. See CONTRIBUTING.md.
Apache 2.0 — free to use, modify, distribute.
Built by NareLabs