adk-go-ollama

Ollama implementation of the model.LLM interface for adk-go, so you can run agents on local models like Llama 3, Mistral, and others with the same ADK APIs you use for Gemini.

Requirements

• Go 1.25+ (aligned with google.golang.org/adk)
• An instance of Ollama running locally or accessible over the network
• golangci-lint if you run make lint (uses .golangci.yaml)

Install

go get github.com/craigh33/adk-go-ollama

Replace the module path with your fork or published path if you rename the module in go.mod.

Makefile

Contributing / Development

Pre-commit hooks

This project uses pre-commit to enforce code quality and commit hygiene. The following tools must be available on your PATH before installing the hooks:

Once the tools are installed, wire the hooks into your local clone:

make pre-commit-install

This installs hooks for both the pre-commit stage and the commit-msg stage.

What the hooks do

Usage

ctx := context.Background()
m, err := ollama.New("gemma3")
if err != nil {
    log.Fatal(err)
}

agent, err := llmagent.New(llmagent.Config{
    Name:  "assistant",
    Model: m,
    Instruction: "You are helpful.",
})
// Wire agent into runner.New(...) as usual.

ollama.New accepts a model name as recognized by your Ollama instance. LLMRequest.Model can override the model name at runtime.

The internal/mappers package holds genai ↔ Ollama conversions (requests, responses, tools, usage). It is internal to this module and used by the ollama package.

Examples

Each example has its own README.md and Makefile:

• examples/ollama-chat: runner-based chat example.
• examples/ollama-tool-calling: tool-calling agent example with function declarations.
• examples/ollama-imagegen: image generation via x/flux2-klein:4b using ADK's tools.
• examples/ollama-stream: direct streaming example using GenerateContent(..., true).
• examples/ollama-multimodal: image analysis and multi-image comparison using vision models like llava.
• examples/ollama-web-ui: ADK local web UI launcher to interact with your Ollama agent.

To run the examples, you can override the default model using the OLLAMA_MODEL environment variable.

make -C examples/ollama-chat run

Run tool calling example:

make -C examples/ollama-tool-calling run

Run streaming example:

make -C examples/ollama-stream run

How it maps to Ollama

• Messages: genai roles user and model map to Ollama user and assistant. System instructions are sent as system messages.
• Tools: the mapper converts GenerateContentConfig.Tools entries (specifically FunctionDeclarations).
• Streaming: When ADK uses SSE streaming, the provider streams the response and yields partial outputs, buffering until the final usage metadata is returned.

Limitations

• Unsupported features: Tool variants not supported by Ollama cause a request-time error. Advanced features that don't map cleanly may be ignored or return an explicit ADK error.
• Multimodal Content: Ollama natively only supports base64-encoded images. While the ADK can handle arbitrary blobs natively (like PDFs, Documents, or Spreadsheets), the adk-go-ollama provider will only process genai.InlineData as images via a vision model. Arbitrary file attachments are ignored or rejected by the Ollama API.
• Image Generation: Ollama's image generation feature currently only supports macOS and x/flux2-klein:4b (based on Flux architectures) and requires a large memory footprint (~12GB). It only supports single image generation requests at a fixed step count natively via /v1/images/generations.

License

Apache 2.0 — see LICENSE.