Tool Calling when using Ollama is Unstable

[mleot]

Pre-flight checklist

• I have searched existing issues and this bug has not been reported yet

Bug description

When I run brv curate with the provider set to locally running ollama models, the output does not call any tools to actually generate files in the context-tree. This is true even when running with the "tested" models for ollama on the docs page (llama3.1:8b, qwen3:8b).

Often the different tools are described by the LLM, but they are not actually executed or used.

This issue does not occur when using LMStudio.

Steps to reproduce

1. Install Ollama, pull models like qwen3 or llama3.1
2. Run brv curate with any content or files

Expected behavior

The model should be able to call tools to do the curation properly.

Actual behavior

No tools are called. Tools or action are described but not called.

CLI version

3.7.1

Installation method

Shell script (curl)

Node.js version

v24.13.1

Operating system

macOS (ARM64 / Apple Silicon)

LLM provider and model

Ollama / llama3.1:8b

Relevant logs

[cuongdo-byterover]

Thanks for the detailed report — this behavior is expected with your current setup.

You're using an outdated recommended model list (from Cipher). The models you mentioned (llama3.1:8b, qwen3:8b) are not reliable for tool calling with Ollama in the current version of brv.

Please refer to the updated recommended LLM list here:
https://docs.byterover.dev/external-llm-providers/overview#recommended-llm-list

If you want to continue using local models with Ollama, switch to a newer model that properly supports tool/function calling, such as:

qwen-3.5-9b (recommended)

Older models may describe tools but fail to actually invoke them — which is exactly what you're seeing.

LMStudio works in your case because it handles tool calling differently and supports a broader set of models more reliably.

Give qwen-3.5-9b a try and it should resolve the issue