add-llm-provider

Add a new LLM provider

First, check the cheap path

If the provider exposes an OpenAI-compatible API (Ollama, Groq, Together, vLLM, Mistral, DeepInfra, Fireworks, Azure OpenAI, …), do not write code. Just set:

export OMNIS_PROVIDER=openai_compat
export OPENAI_BASE_URL=https://api.example.com/v1
export OPENAI_API_KEY=sk-…
export OMNIS_MODEL=their-model-id

Done. Test with go run . console.

If the API is genuinely different, continue.

Implement model.LLM

Add a new file under core/llm/ modelled on anthropic.go or openai.go. Both are ~200 lines, no SDK.

The interface is:

type LLM interface {
    Name() string
    GenerateContent(ctx context.Context, req *model.LLMRequest, stream bool)
        iter.Seq2[*model.LLMResponse, error]
}

Translation primitives are already in core/llm/convert.go — reuse systemTextFromReq, content-part conversions, etc. Don't duplicate.

Register

Edit core/llm/llm.go:

1. Add a default model id to the defaultModel map.
2. Add a case in the New() switch.
3. Document the auth env var(s) at the top of the file.

Test

PATH=$HOME/.local/go/bin:$PATH go build ./... && \
PATH=$HOME/.local/go/bin:$PATH go vet ./... && echo OK

export OMNIS_PROVIDER=newprov
export NEWPROV_API_KEY=…
PATH=$HOME/.local/go/bin:$PATH go run . console
> hello, who are you?

The model's Name() should appear in .agent_events.log's BeforeModel entries.

Update the docs

When the provider works:

1. Add a row to the table in docs/providers.md and the README.
2. If env vars differ from the existing pattern, add a note in .agents/build-and-test/SKILL.md.

Don'ts

• ❌ Don't import a third-party SDK. Use net/http + SSE.
• ❌ Don't change the model.LLM interface — it belongs to ADK.
• ❌ Don't add provider-specific behaviour above the dispatcher; keep the rest of the codebase provider-agnostic.
• ❌ Don't drop streaming support; GenerateContent must honour the stream flag.