How does Hermes Agent pick up Olama?

The key to connecting with Olama is to regard Olama as an OpenAI-compatible custom endpoint. It's not difficult to run. What's really easy to step on is that the context length is too small: the tool schema and system prompts may fill up the 4k context, causing the agent to appear to have "amnesia" or cannot call the tool.

Basic steps

ollama pull qwen2.5-coder:32b
OLLAMA_CONTEXT_LENGTH=32768 ollama serve

and run: 
_CODE_BLOCK_1___1_
Select custom endpoint, fill in the URL_CODE_INLINE_2__, the API key can be skipped, and the model name should be filled in the model name in Ollama.

Check key points

Use_CODE_INLINE_3__to see if CONTEXT reaches 16k or 32k.
Local small models may not be good at complex tool calls, so test with simple file tasks first.
On Windows, make sure that the Olama service can be accessed by WSL or Docker.
In a word: 
Hermes Agent is not difficult to connect with Olama. The difficulty is to provide enough context and choose a model that can stabilize tool calls.
What is local models suitable for?Olama is more suitable for privacy-sensitive, low-cost, light to medium file processing and simple code tasks. If you want to let Hermes Agent plan for a long time, continuously tune tools, and process large warehouses, the reasoning capabilities and context windows of the local model will become bottlenecks. Test with small tasks such as "Read a file and summarize" and "List a directory and modify a text" before using it for formal automation.
Official open source address: https://github.com/NousResearch/hermes-agent; Official document entry: https://hermes-agent.nousresearch.com/_.