agentforge-ollama

agentforge-ollama

Ollama local LLM + embedding provider for AgentForge.

Talks to a local Ollama server (default http://localhost:11434). Use it for fully-local agents, dev loops without an API budget, or to run open-weight models (llama3, qwen3, mistral, mxbai-embed-large) under the same LLMClient / EmbeddingClient contracts as the cloud providers.

Install

pip install "agentforge-ollama[ollama]"

You also need the Ollama daemon running locally:

ollama serve
ollama pull llama3.2:3b

Use

from agentforge import Agent

agent = Agent(model="ollama:llama3.2:3b")

Capabilities

OllamaClient.capabilities() declares {"tools", "streaming"} when the model advertises tool support (the server reports this via /api/show). Caching / extended thinking / vision are not surfaced — Ollama doesn't expose them through a uniform API.

Cost is always zero (local inference).

License

Apache-2.0. See LICENSE.