# go-llm

Go library for chat LLM calls. One client speaks either the Ollama native protocol or an OpenAI-compatible protocol. Callers pass messages. The library does not load files or build product prompts.

Import: `gitea.neitzel.de/konrad/go-llm` (package `gollm`).

```bash
go get gitea.neitzel.de/konrad/go-llm
```

- [Guide](docs/guide.md) — how to call each capability
- [API](docs/api.md) — every exported symbol
- [Domain language](CONTEXT.md)

## Included

- [Adapters](docs/guide.md#adapters) — Ollama native and OpenAI-compatible
- [Chat](docs/guide.md#chat) — one assistant reply, including temperature and max tokens
- [Stream](docs/guide.md#stream) — the same call, delivered as text chunks
- [List Models](docs/guide.md#list-models) — names the server reports

## Excluded

- Embeddings, tools, and raw generate
- Nextcloud types and story types

## Testing

Unit tests use `httptest`. No live model server is required.

`go test ./...` fails when [`docs/api.md`](docs/api.md) does not match the exported API. Regenerate it with:

```bash
UPDATE_API_DOCS=1 go test -run TestAPIDoc -count=1
```

## Related

- **Tales** — ExApp that consumes this library
- Workspace ADR `docs/adr/go-llm/0001-native-adapters.md`
- Workspace ADR `docs/adr/go-llm/0002-wire-only.md`