Documentation: Support connecting to any OpenAI-compatible API
The new [openai] keys, the key rule and the offline demo are described where a user meets them first (README, getting started, deployment) and where a user checks current behaviour (the three capability docs); the docs site builds.
Capability docs
| Capability doc | What changed | History row |
|---|---|---|
configuration.md | [openai] keys and validation (base_url, api, context_window_tokens), OPENAI_API_KEY optional with a base_url, the fail-closed rules | issue-19 |
models.md | endpoint routing (no OPENAI_BASE_URL, no redirects, no org/project headers, keyless), the Chat Completions mapping and finish rule, parse failures as ProviderError, model info and endpoint observability | issue-19 |
demo.md | config.ollama.toml and the Ollama e2e with its skip rule | issue-19 |
Documentation
| Document | What changed |
|---|---|
README.md | "Run the demo": any OpenAI-compatible server via base_url / api, and the offline Ollama run |
guide/getting-started.md | new section "No OpenAI account: any OpenAI-compatible server", with the offline prerequisites and the startup log line |
guide/deployment.md | the [openai] row of the configuration reference and the OPENAI_API_KEY row of the secrets table |
examples/demo/README.md | points to config.embedded.toml and config.ollama.toml for a run with no accounts |
examples/demo/config.ollama.toml | new; its header comments are the bring-up steps |
decisions/decision-006.md | new: the endpoint is configuration-only; a custom endpoint reuses OPENAI_API_KEY |
Site build (T12)
text
$ bun run --cwd docs docs:build
✓ building client + server bundles...
✓ rendering pages...
build complete in 17.25s.