animus-provider-oai

An OpenAI-compatible API provider plugin for Animus.

Status: Under construction — landing in Animus v0.4.x. This crate currently lives in the Animus core workspace at crates/animus-provider-oai/; v0.4.x extracts it to this standalone repository.

What this is

Animus v0.4.0 makes providers (LLM backends) pluggable. This repository will ship animus-provider-oai, a stdio plugin that talks directly to OpenAI-compatible HTTP APIs (the /v1/chat/completions shape) — useful for OpenAI itself, Together AI, Anyscale, on-prem inference servers, vLLM deployments, OpenRouter, and any other OpenAI-shaped backend.

Unlike CLI-wrapper providers (animus-provider-claude, animus-provider-codex, etc.) which spawn an external CLI process per phase, animus-provider-oai makes HTTP calls directly. Useful for backends that don't have a first-class CLI.

Install (planned)

animus plugin install animus-provider-oai

Workflow YAML usage

agents:
  custom-llm:
    model: my-on-prem-llama
    tool: oai
    mcp_servers: ["animus"]

Configure the API endpoint and auth via env vars (OPENAI_API_KEY, OPENAI_BASE_URL, etc.) or per-agent overrides.

Roadmap

• Extract from Animus core workspace at v0.4.x cut
• Publish animus-provider-oai crate to crates.io
• Release binaries on tag
• Independent semver track
• Configurable base URL + auth header per-agent
• CI exercises the contract test from animus-protocol

Design

• Protocol: animus-plugin-protocol (provider variant)
• Naming: repo, crate, and binary all named animus-provider-oai
• Core repo: Animus

License

MIT — see LICENSE.