Agent/Plan mode does not execute tool calls with Ollama remote HTTPS endpoint (RunPod, VS Code Remote SSH)

[wibbycag]

Before submitting your bug report

• I've tried using the "Ask AI" feature on the Continue docs site to see if the docs have an answer
• I'm not able to find a related conversation on GitHub discussions that reports the same bug
• I'm not able to find an open issue that reports the same bug
• I've seen the troubleshooting guide on the Continue Docs

Relevant environment info

## Environment

* **Continue version**: 1.3.30 (linux-x64, installed via VSIX from Open-VSX)
* **VS Code**: Remote SSH
* **OS (remote server)**: Debian GNU/Linux 12 (bookworm)
* **LLM runtime**: Ollama (remote, via HTTPS)
* **Ollama host**: RunPod Pod (GPU A100-SXM4-80GB)
* **API access**: `https://<pod-id>-11434.proxy.runpod.net`
* **Models tested**:

  * `qwen3-coder:30b` (XML tool format)
  * `qwen2.5-coder:32b` (JSON tool format)

---

## Continue configuration (excerpt)


models:
  - name: "Qwen Agent"
    provider: ollama
    model: qwen2.5-coder:32b
    apiBase: https://<pod-id>-11434.proxy.runpod.net
    contextLength: 32768
    roles: [chat]
    capabilities: [tool_use]
    defaultCompletionOptions:
      temperature: 0.15
      topP: 0.9
      maxTokens: 3072


MCPs configured:

* `@filesystem`
* `@tavily`
* `@github`

Description

Summary

Agent and Plan modes do not execute tool calls when using the Ollama provider with a remote HTTPS endpoint (RunPod) in a VS Code Remote SSH environment.
The model generates valid tool calls (JSON or XML), but Continue only displays them as raw output and never executes MCP tools.

Chat mode works correctly.

To reproduce

Steps to reproduce

1. Use VS Code with Remote SSH

2. Install Continue 1.3.30 (VSIX / Open-VSX)

3. Configure Continue with provider: ollama and a remote HTTPS apiBase

4. Enable Agent or Plan mode

5. Ask the agent to use a tool, e.g.:

   Utilise @filesystem pour lire src/example_module/example_file.py
   

Expected behavior

• Tool calls (filesystem, web, git, etc.) are intercepted and executed by Continue
• Tool results are injected back into the model context

Actual behavior

• Model generates valid tool calls (XML or JSON)
• Continue displays the raw tool call output
• No tool is executed
• Same behavior in Agent and Plan modes

Key observations

• Same issue with two different models and two tool formats (XML + JSON)

• Ollama API works correctly when tested directly:

  • /api/tags returns models
  • /v1/chat/completions returns valid OpenAI-style tool calls

• Chat mode works normally

• Switching provider to openai with the same endpoint does not fix the issue

• Suggests a Continue Agent/Plan runtime limitation rather than a model issue

Additional issue: config schema missing

VS Code reports a persistent error:

Problems loading reference 'config-yaml-schema.json': No content

The file config-yaml-schema.json is missing from the VSIX extension directory:

~/.vscode-server/extensions/continue.continue-1.3.30-linux-x64/

This appears related to issues #5545, #6514, #7080.

Question / clarification request

1. Is Agent/Plan mode officially supported with remote Ollama HTTPS endpoints?
2. Is localhost Ollama currently assumed for tool execution?
3. Are there known limitations with VS Code Remote SSH + Ollama?
4. Can debug logs be enabled to understand why tool execution is skipped?

I can provide logs (~/.continue/dev_data/) or a minimal reproduction if helpful.

---

Thank you

Chat mode works great with large coding models.
Agent/Plan modes would unlock the full workflow if remote Ollama were supported.

Log output

Not available.

Continue was fully uninstalled and cleaned after extensive testing.
If needed, I can reinstall and provide fresh logs.

[sestinj]

Related issues found:

Tool execution / Agent mode issues:

• Agent does not execute functions #5696 - Agent does not execute functions (similar symptoms with OpenAI-compatible endpoints)
• Tool / Agent command parsing breaks due to streamed token fragmentation (Ollama + Qwen3-Coder) #9529 - Tool / Agent command parsing breaks due to streamed token fragmentation (Ollama + Qwen3-Coder)

Remote Ollama connection issues:

• Connect error to remote ollama model from Jet Brains IDE #8496 - Connect error to remote Ollama model from JetBrains IDE (ECONNREFUSED 127.0.0.1:80 when using remote apiBase)

Config schema missing (as noted in the issue):

• config.yaml not finding schema #5545 - config-yaml-schema.json missing
• Multiple JSON schema : config-yaml-schema.json #6514 - config-yaml-schema.json error after update
• VS Code extension keeps adding/updating yaml.schemas in settings.json #7080 - Multiple JSON schema issues with config.yaml

[StefanOltmann]

What is the recommended local model that should work?

[github-actions]

This issue hasn't been updated in 90 days and will be closed after an additional 10 days without activity. If it's still important, please leave a comment and share any new information that would help us address the issue.

[Nathaniel-Reeves]

This issue still exists. I was trying this exact setup and experienced the exact bug. It was very disappointing.

[github-actions]

This issue hasn't been updated in 90 days and will be closed after an additional 10 days without activity. If it's still important, please leave a comment and share any new information that would help us address the issue.