Arka

Your terminal, upgraded.
Plain English → 70+ local skills, zero-token routing first,
then a 24-provider LLM failover when you need it.

Docs · Quickstart · PyPI · GitHub

Install

uv tool install "arka-agent[chat]"   # or: pipx install "arka-agent[chat]"
arka setup
arka doctor

One-off without a global install:

uvx --from "arka-agent[chat]" arka doctor

Need the latest commit before the next PyPI release?

pipx install "arka-agent[chat] @ git+https://github.com/Sumit884-byte/arka.git"

Requirements: Python 3.11+ · macOS / Linux full · Windows CLI works (fish unlocks the full router)
Keys: add a free Gemini or Groq key, or run Ollama locally — arka free tier setup helps.

Quick start

arka ask "what is Rust?"
arka "convert 100 USD to INR"
arka council "should I learn Rust?"
arka quiz python
arka listen    # optional: "hey arka, what's the weather"

Connect Cursor / Claude over MCP:

arka mcp doctor && arka mcp install

More: Quickstart · Skills · MCP

Why Arka

How it works

  you type / speak
        │
        ▼
 ┌──────────────────┐     match      ┌─────────────────┐
 │ Symbolic router  │ ─────────────► │ Skill dispatcher │──► skill runs locally
 │ (zero LLM cost)  │                │ + security gate  │
 └────────┬─────────┘                └─────────────────┘
          │ no match
          ▼
 ┌──────────────────┐
 │ LLM failover     │  Gemini · Groq · Ollama · OpenRouter · …
 └──────────────────┘

Same path powers the CLI, MCP server, and remote API. Hosted/headless Linux can set ARKA_HOSTED_MODE=1 to auto-block desktop/GUI/audio skills. Deploy with arka deploy --all (Cloud VM, Railway, Vercel, Netlify, Render).

flowchart TD
    CLI["CLI / NL input"] --> SR["Symbolic router"]
    SR -->|"match"| SD["Skill dispatcher"]
    SR -->|"no match"| LLM["LLM failover chain"]
    LLM --> SD
    SD --> Skills["70+ skills"]
    SD --> MCP["MCP · REST · Telemetry"]

Privacy

• Skills run on your machine — no hosted Arka account or shared demo instance

• Many requests never leave your machine (symbolic routing, zero tokens)

• LLM traffic goes only to providers you configure; force local-only with:

  arka run-only-local-llm "summarize this repo"
  arka hybrid config local-only
  

• Secrets live under your user config (~/.config/arka/ / macOS Application Support / %APPDATA%\arka\)

• Memory defaults local; web content is sanitized; risky actions prompt [y/N]

• Telemetry defaults to local SigNoz (127.0.0.1:4318) — disable with OTEL_SDK_DISABLED=true

Details: Security · Memory

Platforms

Develop from source

git clone https://github.com/Sumit884-byte/arka.git
cd arka
./scripts/refetch.sh --install
pip install -e ".[chat,dev]"
arka setup && arka doctor

Fork + PR: see CONTRIBUTING.md. Local landing preview: landing/.

Contributing

Issues and PRs welcome — start with good first issue.

gh repo fork Sumit884-byte/arka --clone
cd arka && pip install -e ".[chat,dev]" && pytest
gh pr create --repo Sumit884-byte/arka

If Arka is useful, ★ star the repo — or gh repo star Sumit884-byte/arka.

License

MIT