A terminal coding agent

Ship code from your terminal with stella

A fast, bring-your-own-key, model-agnostic coding agent. Point it at any provider, give it a goal, and let a verifier decide when the work is actually done.

zsh — stella

$ export ANTHROPIC_API_KEY=…

$ stella run "fix the failing test"

Bring your own key

No account, no sign-up. Stella auto-detects the provider from whichever API keys you have set and runs on your credentials.

Model-agnostic

Anthropic, OpenAI, Gemini, Vertex, Bedrock, xAI, DeepSeek, Z.ai, OpenRouter, and any OpenAI-compatible local server — one CLI.

Verified outcomes

Goal mode doesn't stop on a hunch. A separate judge model verifies the definition of done from evidence before the loop ends.

Real tools

Read, edit, glob, grep, shell, CI, screenshots, issues, and developer-defined script tools — with a per-tool permission model.

MCP native

Connect Model Context Protocol servers from .stella/mcp.toml to bring more tools into the agent at session start.

Local telemetry

Token usage, cost, and step metering land in a local SQLite store on your disk — inspect every run, share nothing.

Works with your provider

AnthropicOpenAIGoogle GeminiVertex AIAmazon BedrockxAIDeepSeekZ.aiOpenRouterLocal / OpenAI-compatible

Override any provider's base URL, key, or model in settings.json — no provider-specific environment variables required.

Goal mode

Give Stella an objective and it drives to green — editing, running, and re-checking until a judge confirms the goal from evidence.

How goal mode works

Credentials & scopes

A clear credential chain and a three-tier settings hierarchy — project, org-managed, and user — so teams share defaults safely.

Configure settings

Extend it

Add MCP servers, custom script tools, skills, and lifecycle hooks. Stella meets your workflow instead of replacing it.

Add your own tools