Stella overview

stella is an open-source coding agent that runs in your terminal. You give it a task in natural language and it reads your code, edits files, runs commands, and iterates until the work is done. Where most agents stop at "the suite is green," Stella holds itself to a harder standard: a witness test that fails on your old code and passes on the new, checked by an LLM judge with its own read-only tools.

Stella is built by the team behind Oxagen, but it stands alone. There is no account, no sign-up, and no platform dependency. It is bring-your-own-key: set an API key for any of nine supported providers, or point it at a local model server and use no key at all. The only network traffic Stella produces is to the model provider you chose.

The source lives at github.com/oxageninc/stella-cli, dual-licensed MIT OR Apache-2.0.

Getting started in three commands

# 1. Install (requires Rust 1.90+, see the installation guide)
cargo install --locked --git https://github.com/oxageninc/stella-cli stella-cli

# 2. Set a key for any supported provider
export ANTHROPIC_API_KEY=your_key_here   # or OPENAI_API_KEY, GEMINI_API_KEY, …

# 3. Start working
stella run "fix the failing test in src/auth.rs"

From there, stella with no arguments opens an interactive REPL, and stella goal works in judged rounds until the goal is verifiably met.

Why Stella

• Verified done, not claimed done. The verify_done gate replays your witness test against the previous code in a shadow worktree. It must fail there and pass on your change. WITNESS CONFIRMED is the only accepted finish line; a merely green suite is not.
• Any model, your key. Anthropic, OpenAI, Google Gemini, Vertex AI, Amazon Bedrock, xAI, DeepSeek, Z.ai, OpenRouter, or any OpenAI-compatible local server (Ollama, vLLM, LM Studio, llama.cpp). See Models and providers.
• A hard dollar budget. --budget 5 aborts the run cleanly (never mid-tool) once total spend crosses $5.00. Without it, spend is still metered and reported after every turn.
• Local-first telemetry. Every execution is recorded in .stella/stella.duckdb in your workspace: the full event stream, per-call token and cost telemetry, and the files-touched ledger. Query it with any DuckDB client. Nothing leaves your machine.
• Self-improving. Lessons saved with save_memory (or written by you as markdown in .stella/memories/) load into a byte-stable system prompt at session start, so every call reads them at prompt-cache-hit prices.

Who this is for

These docs are written for developers. You should be comfortable with a terminal, environment variables, and a Rust toolchain (only for installation — using Stella requires no Rust knowledge). No Oxagen account or platform knowledge is needed anywhere in these pages.

Where next

• Installation — cargo install, building from source, updating, uninstalling.
• Quickstart — first chat, first one-shot run, first judged goal.
• Commands — every subcommand, flag, REPL command, and built-in tool.
• Models and providers — the full provider matrix and how auto-detection picks one.

Stella and the Oxagen CLI

Stella is not the oxagen CLI. The oxagen CLI is a client for the Oxagen platform: it authenticates against your organization, queries your workspace knowledge graph, and meters usage through the platform. Stella is a standalone open-source coding agent with no platform attached. If you want a terminal agent grounded in your organization's knowledge graph, use the oxagen CLI; if you want a fast, self-contained, bring-your-own-key coding agent in any repo, use Stella.