docker-infrastructure/openclaw/TROUBLESHOOTING.md

# OpenClaw Troubleshooting Log

## Issue Summary
OpenClaw gateway mode consistently fails with "No API key found for provider 'anthropic'" despite correct configuration of auth-profiles.json with both Ollama and Anthropic API keys.

**Date**: 2026-02-09
**OpenClaw Version**: `ghcr.io/openclaw/openclaw:main` (2026.2.6-3)
**Environment**: Docker on x86_64 Linux

## Errors Encountered

### 1. Architecture Mismatch (SOLVED)
**Error**: `[FATAL tini (7)] exec docker-entrypoint.sh failed: Exec format error`

**Cause**: Docker image defaults to ARM64 architecture

**Solution**: Add `platform: linux/amd64` to docker-compose.yml

```yaml
services:
  openclaw:
    platform: linux/amd64
```

### 2. Device Pairing Required (SOLVED)
**Error**: `disconnected (1008): pairing required`

**Solution**: Approve pending device pairing requests
```bash
docker exec openclaw npx openclaw devices list
docker exec openclaw npx openclaw devices approve <REQUEST_ID>
```

### 3. API Key Not Found (UNSOLVED)
**Error**:
```
Error: No API key found for provider "anthropic".
Auth store: /home/node/.openclaw/agents/main/agent/auth-profiles.json
```

**This error persists despite all configuration attempts below.**

## Configuration Attempts

### Attempt 1: Ollama Integration via Environment Variables
Added environment variables to docker-compose.yml:
```yaml
environment:
  - OPENCLAW_AGENT_PROVIDER=ollama
  - OPENCLAW_AGENT_MODEL=llama3
  - OPENCLAW_OLLAMA_BASE_URL=http://ollama:11434
```
**Result**: ❌ Gateway still defaulted to Anthropic

### Attempt 2: Main Config with Ollama
Created `/srv/openclaw/config/config.json`:
```json
{
  "agents": {
    "defaults": {
      "provider": "ollama",
      "model": "llama3",
      "authProfile": "ollama"
    }
  },
  "ollama": {
    "baseURL": "http://ollama:11434"
  }
}
```
**Result**: ❌ Gateway ignored configuration

### Attempt 3: OpenAI-Compatible Ollama Configuration
Based on GitHub issues showing Ollama uses OpenAI-compatible API:
```json
{
  "models": {
    "providers": {
      "ollama": {
        "baseUrl": "http://ollama:11434/v1",
        "api": "openai-completions",
        "models": ["llama3", "qwen3-coder"]
      }
    }
  }
}
```
**Result**: ❌ Gateway still tried to use Anthropic

### Attempt 4: Dummy Anthropic Token for Ollama Redirect
Created auth-profiles.json with anthropic profile pointing to Ollama:
```json
{
  "anthropic": {
    "provider": "openai",
    "baseURL": "http://ollama:11434/v1",
    "apiKey": "sk-ollama-dummy"
  }
}
```
**Result**: ❌ Still complained about missing Anthropic API key

### Attempt 5: Real Anthropic API Key (Simple Format)
```json
{
  "anthropic": {
    "apiKey": "sk-ant-api03-..."
  }
}
```
**Result**: ❌ Same error

### Attempt 6: Provider:Profile Name Format
Based on GitHub issues showing profile names like "anthropic:claude-cli":
```json
{
  "anthropic:default": {
    "provider": "anthropic",
    "apiKey": "sk-ant-api03-..."
  }
}
```
**Result**: ❌ Same error

### Attempt 7: Multiple File Locations
Tried placing auth-profiles.json in:
- `/srv/openclaw/config/auth-profiles.json` (root level)
- `/srv/openclaw/config/agents/main/agent/auth-profiles.json` (agent level)
- Both locations simultaneously

**Result**: ❌ Same error regardless of location

### Attempt 8: Agent-Specific Config
Created `/srv/openclaw/config/agents/main/agent/config.json`:
```json
{
  "provider": "openai",
  "model": "llama3",
  "authProfile": "anthropic",
  "openai": {
    "baseURL": "http://ollama:11434/v1"
  }
}
```
**Result**: ❌ Same error

### Attempt 9: Ollama Launch Command
Tried using Ollama's built-in OpenClaw launcher:
```bash
docker exec ollama ollama launch openclaw --config --model llama3
```
**Result**: ✅ Command succeeded and said "Added llama3 to OpenClaw"
**But**: Gateway mode in Docker still couldn't use it

## Files Verified

All configurations were verified to be:
- ✅ Valid JSON (validated with `python3 -m json.tool`)
- ✅ Correct permissions (`chown 1000:1000`, owned by `node` user)
- ✅ In correct locations (verified with `docker exec` inside container)
- ✅ Properly mounted via Docker volumes

## Network Verification

Confirmed OpenClaw and Ollama connectivity:
```bash
docker exec openclaw curl -s http://ollama:11434/api/tags
# Successfully returned list of available models including llama3
```

Both containers on same network (`npm-network`): ✅

## Key Findings

### 1. Gateway Mode Default Behavior
The log message `agent model: anthropic/claude-opus-4-6` appears on every startup regardless of configuration, suggesting gateway mode with `--allow-unconfigured` flag has hardcoded Anthropic as default.

### 2. Ollama Integration Method
Official documentation shows `ollama launch openclaw` is intended for HOST installation, not containerized deployment. The containerized gateway mode may not support Ollama properly.

### 3. Auth Profile Format
GitHub issues show various profile formats:
- `"anthropic:claude-cli"` with `"mode": "oauth"`
- `"ollama:local"` with `"token": "..."`
- `"minimax-cn:default"` with `"provider": "minimax"`

None of these formats worked in our containerized setup.

## Documentation Links

- **OpenClaw GitHub**: https://github.com/openclaw/openclaw
- **Ollama Integration Docs**: https://docs.ollama.com/integrations/openclaw
- **OpenClaw + Ollama Setup Guide**: https://openclawai.net/blog/openclaw-ollama-setup
- **GitHub Discussion #2936**: https://github.com/openclaw/openclaw/discussions/2936 (Ollama usage)
- **GitHub Issue #3740**: https://github.com/openclaw/openclaw/issues/3740 (Ollama auth check)
- **GitHub Issue #1253**: https://github.com/openclaw/openclaw/issues/1253 (Auth profile format)

## Current Working Configuration

### docker-compose.yml
```yaml
services:
  openclaw:
    container_name: openclaw
    image: ghcr.io/openclaw/openclaw:main
    platform: linux/amd64
    restart: unless-stopped
    command: ["node", "openclaw.mjs", "gateway", "--allow-unconfigured", "--bind", "lan"]
    environment:
      - HOME=/home/node
      - TERM=xterm-256color
      - TZ=America/New_York
      - PUID=1000
      - PGID=1000
      - OPENCLAW_GATEWAY_TOKEN=27d4e63adce6c8f7c5396e8ca3f9ec5e6ff590077247fb11da03a8684ee3c711
      - OPENCLAW_GATEWAY_BIND=lan
      - VIRTUAL_HOST=openclaw.kolpacksoftware.com
      - VIRTUAL_PORT=18789
      - LETSENCRYPT_HOST=openclaw.kolpacksoftware.com
    volumes:
      - /srv/openclaw/config:/home/node/.openclaw
      - /srv/openclaw/workspace:/home/node/.openclaw/workspace
    init: true
    networks:
      - npm-network

networks:
  npm-network:
    external: true
```

### Status
- ✅ Container runs without crashes
- ✅ Gateway accessible at https://openclaw.kolpacksoftware.com
- ✅ WebSocket connections work
- ✅ Device pairing works
- ❌ Cannot use any AI model (Ollama or Anthropic)

## Possible Causes

1. **Docker Image Issue**: The containerized version may have a bug or missing configuration
2. **Gateway Mode Limitation**: `--allow-unconfigured` flag may bypass auth-profiles.json entirely
3. **Version-Specific Bug**: This version (2026.2.6-3) may have a regression
4. **Missing Setup Step**: There may be an initialization step we're missing
5. **Auth System Design**: The auth system may be designed for host installation only

## Recommended Solutions to Try

### 1. Install on Host Instead of Docker
```bash
npm install -g openclaw@latest
openclaw onboard --install-daemon
ollama launch openclaw
```
This follows the official documentation more closely.

### 2. Try Different Docker Image Tag
```yaml
image: ghcr.io/openclaw/openclaw:2026.2.5  # or other specific version
```

### 3. Remove `--allow-unconfigured` Flag
May require running `openclaw onboard` inside container first (interactive, difficult in Docker).

### 4. File GitHub Issue
Report this configuration with:
- docker-compose.yml
- auth-profiles.json attempts
- Full error logs
- Ask maintainers for Docker-specific setup guide

### 5. Alternative AI Assistants
- **Open WebUI**: Excellent Ollama integration, easy Docker setup
- **LibreChat**: Multi-provider support including Ollama
- Both are well-documented and Docker-friendly

## Next Steps

Before revisiting:
1. Check OpenClaw GitHub for updates/fixes related to Docker deployment
2. Look for new documentation about auth-profiles.json format
3. Search for successful Docker deployment examples
4. Consider whether host installation is more appropriate for this use case

## Environment Details

- **Host OS**: Linux 6.8.0-94-generic (Ubuntu 24.04)
- **Docker Version**: 28.2.2
- **Architecture**: x86_64
- **Ollama Version**: Running in separate container on same network
- **Ollama Models**: llama3:latest, qwen3-coder, qwen3:8b, gemma3:4b, and others
- **Network**: npm-network (shared with nginx-proxy)