Goose

Local AI agent — Ollama, multi-provider, full-auto mode

Run a full AI coding agent on local hardware. Goose connects to Ollama for local inference. No API keys, no cloud dependency, no data leaving the machine. When cloud power is needed, switch to Anthropic, OpenAI, or Google with one setting change.

Goose configuration

Local inference with Llama

Agent slot configuration

Ollama local Qwen chat

Configure provider, model, and runtime settings — one panel for local and cloud inference.

Capabilities

What it does

Local Inference via Ollama

Run open-weight models on local hardware. No API keys, no usage fees, no data leaving the machine. Ollama manages model downloads, quantisation, and GPU acceleration.

Multi-Provider Fallback

Switch between Ollama (local), Anthropic, OpenAI, Google, Groq, Mistral, Bedrock, Azure, and Databricks. Use local models for privacy-sensitive work, cloud models for more capability.

Full-Auto Mode

Run Goose in full-auto mode — no approval prompts, no human-in-the-loop pauses. The agent executes its entire plan autonomously. Ideal for bulk operations, migrations, and test generation.

ACP Protocol

Communicates via Agent Communication Protocol (ACP) over JSON-RPC 2.0. Structured message passing with streaming support — not raw stdin/stdout parsing.

Per-Slot Configuration

Each agent slot can use a different Goose provider and model. Run the security reviewer on a local model for privacy, and the coder on Claude for capability, in the same workspace.

Health Monitoring

Setup wizard validates the configuration. Status page shows process state, model loaded, and connection health. Automatic restart on crash with configurable retry policy.

How it works

From install to first use.

Choose the providerThe setup wizard guides through provider selection. Choose Ollama for local inference (downloads and manages models) or configure a cloud provider with API keys.

Assign to agent slotsEach specialist agent in the workspace can use Goose with a different configuration. The privacy-focused agent runs local; the capability-focused agent runs cloud.

Agent works autonomouslyIn full-auto mode, Goose executes without approval prompts. It reads files, writes code, runs tests, and iterates, all within the configured boundaries.

Switch providers any timeChange from local to cloud (or back) with one setting change. No workflow changes, no reconfiguration. The agent's identity and history persist.

Why local matters

Your inference, your hardware, your choice.

Local inference via Ollama means zero data leaves the machine. When cloud capability is needed, the team chooses which provider and which model, not the platform.

0data sent to cloud with Ollama

9+providers — local and cloud

Freelocal inference with open-weight models

1 clickto switch between local and cloud

Goose ships with the Studio. No extra install, no extra cost.

Get early access All extensions