Troubleshooting¶
This guide covers the most common errors encountered when running Neuralgentics.
🚨 Common Errors¶
1. "Kimi K2.6 not valid" / Model Not Found¶
Symptom: The TUI crashes or the orchestrator fails to dispatch with a ProviderModelNotFoundError.
Root Cause: Usually a mismatch between the model name in .opencode/agents/*.md and the models list in .opencode/opencode.json. OpenCode requires exact string matches.
Fix: 1. Open .opencode/opencode.json. 2. Ensure the ollama provider block contains a models array. 3. Verify the model name is exactly kimi-k2.6 (without the :cloud suffix if using the Ollama API directly). 4. Restart OpenCode to clear the config cache.
2. Broker 401 / Unauthorized¶
Symptom: An agent receives unauthorized: role "writer" cannot access server "github-mcp".
Root Cause: The agent is attempting to use a tool it is not permitted to use according to the Routing Matrix.
Fix: - If the agent should have access, update packages/broker-go/src/neuralgentics/broker/access/access.go. - If it's a routing error, the Orchestrator must be corrected to route the task to boomerang-git.
3. Memini-AI Down / Connection Refused¶
Symptom: failed to connect to memory server at localhost:8000.
Root Cause: The Python memory server is not running or the PostgreSQL backend has crashed.
Fix:
Verify the DB is healthy:docker ps | grep postgres. 4. Kanban Card "Stuck" in Running¶
Symptom: A card remains in running status but the agent has stopped responding.
Root Cause: A sub-agent crashed or the session timed out without a wrap-up.
Fix: 1. Issue the /resume command in the TUI. 2. This resets the failure counter and moves the card back to ready. 3. The orchestrator will re-dispatch the task.
🛠️ Debugging Commands¶
| Command | Purpose |
|---|---|
neuralgentics status | Check connectivity to Backend and Memory. |
curl http://localhost:8000/status | Check health of the Python memory engine. |
make lint | Verify Go source code quality. |
make smoke | Run the JSON-RPC integration test. |