Ollama Not Working - Session Locked Out

[obsidian harbor]

I have the following configuration:

{
  "agents": {
    "defaults": {
      "maxConcurrent": 4,
      "subagents": {
        "maxConcurrent": 8
      },
      "compaction": {
        "mode": "safeguard"
      },
      "workspace": "/home/node/clawd",
      "model": {
        "primary": "ollama/mistral-large-3:675b-cloud"
      }
    }
  },
  "models": {
    "providers": {
      "ollama": {
        "baseUrl": "http://redacted.com:8081/api",
        "apiKey": "REDACTED",
        "api": "ollama", // I have also tried openai and openai-completions here
        "models": [
          { "id": "devstral-2:123b-cloud", "name": "Devstral 2", "reasoning": true },
          { "id": "mistral-large-3:675b-cloud", "name": "Mistral Large 3", "reasoning": true },
          { "id": "deepseek-v3.2:cloud", "name": "DeepSeek 3.2", "reasoning": true }
        ]
      }
    }
  }
}

If I use "ollama" for the "api" configuration parameter, I get an error on startup. If I use openai-completions, I get a 404 error when I send a message to my bot. If I use openai, I get no error or log on the ollama side, and no response from the bot. It then locks me out of my session

clawdbot-gateway-1  | 2026-01-26T16:07:56.871Z [diagnostic] lane task error: lane=main durationMs=10543 error="Error: session file locked (timeout 10000ms): pid=7 /home/node/.clawdbot/agents/main/sessions/a149e1b3-d694-455b-8390-b77061161c16.jsonl.lock"

[grand viper]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[jovial urchin]

Now I understand the issue. Let me provide the solution:

The problem: The Ollama configuration has two critical issues:

1. Missing /v1 in baseUrl - Ollama's OpenAI-compatible API requires the /v1 endpoint
2. Wrong api setting - api: "ollama" doesn't exist; use implicit discovery or explicit with openai-completions

Fix your config to:

{
  models: {
    providers: {
      ollama: {
        // Must include /v1 for OpenAI-compatible API
        baseUrl: "http://redacted.com:8081/v1",

        apiKey: "ollama-local",  // Any value works for Ollama
        api: "openai-completions",
        models: [
          { 
            "id": "mistral-large-3:675b-cloud",  // Use exact Ollama model name
            "name": "Mistral Large 3",
            "reasoning": true 
          },
          { 
            "id": "deepseek-v3.2:cloud",
            "name": "DeepSeek 3.2",
            "reasoning": true 
          }
        ]
      }

    }
  },
  agents: {
    defaults: {
      model: {
        primary: "ollama/mistral-large-3:675b-cloud"
      }
    }
  }
}

Or even simpler - use implicit discovery:

If your remote Ollama is accessible, just set environment variable and omit explicit config:

export OLLAMA_API_KEY="ollama-local"
# Don't define models.providers.ollama at all

Clawdbot will auto-discover models from http://redacted.com:8081 (adds /v1 and /api/tags internally).

Key points:

• baseUrl MUST end with /v1 for explicit config
• Use api: "openai-completions" (not "ollama" or "openai")
• Model IDs must match exactly what ollama list shows
• Port 8081 is fine, just needs the /v1 suffix

[obsidian harbor]

Now I get an HTTP 500 reply from the bot with no errors in my ollama logs

[jovial urchin]

HTTP 500 with no Ollama logs means the request is reaching Ollama but failing. Let's debug:

First, test your Ollama API directly:

# List models on remote server
curl http://redacted.com:8081/api/tags

# Try a simple completion test
curl http://redacted.com:8081/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "mistral-large-3:675b-cloud",
    "messages": [{"role": "user", "content": "hi"}],
    "max_tokens": 50
  }'

Common issues:

1. Model name format - Remote Ollama might need a different name format. What does curl /api/tags return?

2. Model not pulled - Is mistral-large-3:675b-cloud actually available on that remote Ollama instance?

3. Try without :cloud suffix - Some remote Ollama setups use just the model name:

   "model": "mistral-large-3"
   

4. Check Ollama server logs on the remote host - HTTP 500 should appear there.

Quick config to test - simplify the model ID:

{
  models: {
    providers: {
      ollama: {
        baseUrl: "http://redacted.com:8081/v1",
        apiKey: "ollama-local",
        api: "openai-completions",
        models: [
          { 
            "id": "mistral-large-3",  // Try without :675b-cloud
            "name": "Mistral Large 3",
            "reasoning": true 
          }
        ]
      }

    }
  }
}

What does curl http://redacted.com:8081/api/tags return on the remote Ollama?

[obsidian harbor]

It returns JSON with a list of my local models. I also tried the "simple completion test" and it worked successfully - I got JSON with a valid response back

If I try omitting the :suffix I get "model 'mistral-large-3' not found" as an error from the bot

[jovial urchin]

The API works fine, so the 500 is coming from Clawdbot, not Ollama. Check your gateway logs for the actual error:

# Get full gateway logs with the 500 error
clawdbot gateway logs --tail=100 | grep -i "500\|error\|mistral"

Also check: The session lock error from earlier might be corrupting things. Try:

# Clear stuck locks
rm -f ~/.clawdbot/agents/main/sessions/*.lock

# Restart gateway
clawdbot gateway restart

# Then test again

Most likely causes:

1. Context window issue - Add explicit context settings:

   {
     "models": {
       "providers": {
         "ollama": {
           "models": [
             {
               "id": "mistral-large-3:675b-cloud",
               "name": "Mistral Large 3",
   

               "reasoning": true,
               "contextWindow": 131072,
               "maxTokens": 131072
             }
           ]
         }
       }
     }
   }

2. Cost settings - Explicit costs can help:

   "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
   

What does clawdbot gateway logs show when you send a message?

[obsidian harbor]

How do I execute that using docker compose?

Context window and cost settings made no difference, still getting a 500

[jovial urchin]

# View gateway logs
docker compose logs clawdbot-gateway -tail=100

# Or follow logs in real-time
docker compose logs -f clawdbot-gateway

# Clear locks inside the container
docker compose exec clawdbot-gateway rm -f /home/node/.clawdbot/agents/main/sessions/*.lock

# Restart the gateway service
docker compose restart clawdbot-gateway

Quick one-liner to see errors:

[jovial urchin]

We need to see the actual error. Run:

docker compose logs clawdbot-gateway 2>&1 | tail -50

Paste the output here - the 500 error details will tell us exactly what's failing.

docker compose logs clawdbot-gateway 2>&1 | grep -i "error\|500\|mistral\|failed"