Launch imminent: Warden v0.1.0
A lean, multi-provider, terminal-native LLM coding agent. Single binary. MIT. Pre-Alpha ships soon.
Something is coming: a terminal-native LLM coding agent that doesn’t lock you
to one provider. Anthropic Claude, OpenAI (Platform API plus the Responses
subscription tier), and Google Gemini are all fully wired today. Bedrock,
Vertex, and Foundry are scaffolded for tomorrow, each raising an explicit
NotImplementedError so you can’t reach for one by accident.
The surface you get on day one: twenty-one builtin tools, a full MCP client and server mode, six permission modes governed by an eight-source precedence chain, and twenty-two hook event types. It’s Python 3.14 in a single binary, MIT licensed, and the internal architecture is layered strictly enough that a sideways dependency can’t sneak in and rot the codebase.
Telemetry is off by default. Turn it on when you want full diagnostics, kill
it with WARDEN_TELEMETRY=off, and know that what it can send is
schema-allowlisted server-side. License activation talks to a small
first-party endpoint at api.wardenclient.com, and the full data contract is
written out in docs/telemetry.md in the repository.
We don’t promise vaporware. The Pre-Alpha tag is honest about where the edges are: three providers fully wired, three scaffolded with explicit error placeholders, and every known gap documented on the homepage of the README itself. What ships, ships.
Was this useful?