Continue.dev does not use tools

[didiator]

Hi,
hopefully someone can help me. I want to work offline with my graphics card (28 GB VRAM) using QWen3 coder 4k quantization. The model seems to load well (see nv
nvidia-smi.log
idia-smi.log).
I built the model using the following modified model file:
Modelfile.txt
My config.yaml is this:
config.yaml.txt

The model, however, finds very few tools internally. For example, if I ask: Where in file XY is class YZ instantiated, I get the answer:
I understand you want me to show you the exact location in main.py where the wake word detection is initiated. I have to be honest—I can't directly access the file to read the code.
...
You'll have to search through the code yourself.

I cannot understand this. What causes this behavior and what can I do against this?

VS Code version: 1.109.4
continue.dev version: 1.2.16
ollama version is 0.16.1
QWen3-coder

[xXMrNidaXx]

Tool usage issues are often config-related! At RevolutionAI (https://revolutionai.io) we use Continue.dev extensively.

Common fixes:

1. Check model supports tools:

{
  "models": [{
    "title": "GPT-4",
    "provider": "openai",
    "model": "gpt-4-turbo",
    "capabilities": ["tools"]
  }]
}

2. Verify tool definitions:

{
  "tools": [{
    "name": "search",
    "description": "Search the web",
    "parameters": { ... }
  }]
}

3. Enable in context:
   Some models need explicit tool-use prompting in the system message.

4. Check provider compatibility:

• OpenAI: ✅ Full support
• Anthropic: ✅ Full support
• Ollama: ⚠️ Model-dependent
• Local: ⚠️ Often limited

Debug approach:

• Check logs for tool call attempts
• Try a simpler tool first (calculator)
• Verify JSON schema is valid

What model and provider are you using?

[xXMrNidaXx]

The model is not using tools because it needs explicit tool-calling configuration.

Issue:
Qwen3-coder needs to be configured to use tools/functions, not just chat.

Fix 1: Enable tools in config

models:
  - name: Qwen3-coder
    provider: ollama
    model: qwen3-coder:4k
    capabilities:
      - tools  # Enable tool calling
    toolCallStrategy: auto

Fix 2: Use Continue tools explicitly

tools:
  - type: codebase
    enabled: true
  - type: file
    enabled: true
  - type: search
    enabled: true

Fix 3: Check model supports tools

Not all models support function calling. Try:

# Test tool support
ollama run qwen3-coder:4k
>>> /tools

Fix 4: Use @ mentions

Continue uses @ mentions for file access:

@main.py Where is class YZ instantiated?

Fix 5: Index your codebase

Ensure Continue has indexed your workspace:

• Open command palette
• Run "Continue: Index Codebase"

Alternative model:
If Qwen3 does not support tools well, try:

• deepseek-coder-v2
• codellama:34b
• qwen2.5-coder:32b

We optimize Continue.dev setups at Revolution AI — tool configuration + codebase indexing is usually the fix.

[f1-outsourcing]

The model is not using tools because it needs explicit tool-calling configuration.

Issue: Qwen3-coder needs to be configured to use tools/functions, not just chat.

Fix 1: Enable tools in config

models:
  - name: Qwen3-coder
    provider: ollama
    model: qwen3-coder:4k
    capabilities:
      - tools  # Enable tool calling
    toolCallStrategy: auto

where did you see tools???? That does not exist
https://docs.continue.dev/reference

[ibondarenko1]

the previous answers don't apply to your stack, you're on ollama not openai, and capabilities: [tools] in continue's config is more of a hint than an enforcer.

three things to verify, in this order:

does your local qwen3-coder GGUF actually have a tool-calling chat template? a lot of community quants strip it. check:

ollama show qwen3-coder:4k --modelfile | grep -A 30 TEMPLATE

if you don't see jinja conditionals around tools (something like {%- if tools %}), the model can't emit tool calls regardless of what continue sends. you need a different GGUF or to bake the template back in via Modelfile.

does ollama think the model supports tools? in recent ollama versions:

ollama show qwen3-coder:4k

look for Capabilities: ... tools. if tools isn't listed, the ollama side won't surface them to continue's MCP/tools layer no matter how you configure capabilities on the continue side.

is your continue config sending tools through? continue 1.2.x for ollama models the right shape is roughly:

models:
  - name: qwen3-coder
    provider: ollama
    model: qwen3-coder:4k
    apiBase: http://localhost:11434
    roles:
      - chat
      - edit
    capabilities:
      - tools

then check the continue dev console (cmd+shift+P > "Continue: Toggle Developer Tools") with verbose logging on, send a message, and watch for the tools: [...] block in the request continue makes to ollama. if that block is empty or missing, the issue is on the continue side. if the block is there but the model still answers "i can't access the file", it's the model template (point 1).

the "i can't access the file" reply is classic for a model that has the tool definitions in context but no template hook to actually invoke them.

[didiator]

Hi, Thank you for your replies. In the meantime i did find another approach. Non of that solutions did work well.

[f1-outsourcing]

I have the impression it is also quite buggy and I wonder if this is the models changing communication or the continuedev plugin. Sometimes I have lots of file access going right and then it is failing constantly. I also wonder if these rules are send because these models do not seem to use them.