Opencode + dockerized ollama getting stuck after first prompt

[lopeztel]

Description

Running docker container with ollama, managing through openwebui.

OS: Bazzite
Opencode version: 1.14.31

Problem:
I have tried to use both qwen3.5:9b and qwen3.5:4b locally with the same results, the first prompt requires running command ls in the current directory. After reasoning that ls must be ran, the model stops execution fully without executing anything, furthermore all context is forgotten.

Expected behavior:
running commands in the current directory and being able to remember the context.

Settings:

• I have ollama deployed as per instructions here
• Based on instructions here, this is my config:

{
  "$schema": "https://opencode.ai/config.json",
  "default_agent": "plan",
  "compaction": {
    "auto": true,
    "prune": true,
    "reserved": 8192
  },
  "provider": {
    "ollama": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Ollama (local)",
      "options": {
        "baseURL": "http://localhost:11434/v1"
      },
      "models": {
        "qwen3.5:9b": {
          "name": "qwen3.5:9b_ext",
          "_launch": true,
          "tool_call": true,
          "options":{
            "temperature": 0.1
          },
          "maxTokens": 16384
        },
        "qwen3.5:4b": {
          "name": "qwen3.5:4b_ext",
          "_launch": true,
          "tool_call": true,
          "options":{
            "temperature": 0.1
          },
          "maxTokens": 16384
        },
        "gemma4:e4b": {
          "name": "gemma4:e4b_ext",
          "_launch": true,
          "tool_call": true,
          "options":{
            "temperature": 0.1
          },
          "maxTokens": 16384
        }
      }
    }
  },
  "permission":{
    "read": "allow"
  }
}

• I have modified parameter num_ctx in openwebui to be 65536:

I'm happy to help with more info if needed. Feel free to point me in the right direction if I missed something

Plugins

No response

OpenCode version

1.14.31

Steps to reproduce

1. deploy ollama in docker (AMD rocm)
2. deploy openwebui in docker to download models (qwen3.5:4b used in example but 9b has the same problem)
3. in openwebui, go to workspace->new model->select base model (previously downloaded qwen3.5:4b)->advanced params -> show, modify num_ctx = 65526
4. Modify opencode config json
5. run opencode in any directory, select the corresponding model, change to build, ask it to do ls in the current directory or list its contents

Screenshot and/or share link

Operating System

Bazzite

Terminal

Konsole

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• OpenCode 1.4.3 hangs with local Ollama provider on simple prompts, while /v1/chat/completions works #22132: OpenCode hangs with local Ollama provider on simple prompts — very similar symptom (model reasons but execution stops/hangs), same qwen3.5 model family and @ai-sdk/openai-compatible setup
• Thinking but no answer or action when using local models #20906: Thinking but no answer or action when using local models — nearly identical setup (Ollama + Open WebUI) with the same symptom of the model reasoning but not executing tool calls
• Ollama context stuck at 0% #21736: Ollama context stuck at 0% — same qwen3.5 model via Ollama, similar stuck/halted behavior

If your issue is the same as one of these, please add a comment there with any additional details (e.g. the AMD ROCm / Docker-specific angle may be a useful addition).

[lopeztel]

Trying a configuration like this one didn't work either

[github-actions]

To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.