Small local testing
Use compact instruct models for chat, settings validation, and simple tool calls.
Models and Cookbook
Pick the right model before judging Odysseus.
A weak or mismatched local model can make any agent workspace feel broken. Start with a model that fits your memory, context needs, and latency tolerance.
Rule of thumb
Use smaller models first to validate the app. Increase model size only after chat, tools, and memory are stable.
Balanced agents
Try stronger 7B to 14B class models for planning, summaries, and coding help.
Heavy workflows
Use larger models for deeper reasoning, longer context, and multi-step agent tasks.
Ollama setup
Connect Odysseus to a local model server
Ollama is a common starting point because it simplifies pulling and serving local models. After pulling a model, confirm it appears in Ollama before debugging Odysseus.
ollama pull qwen2.5:7b
ollama run qwen2.5:7b
ollama listIf the model does not show inside Odysseus, check the endpoint URL, container networking, and whether Odysseus is running inside Docker while Ollama is running on the host.