07 · Troubleshooting
The most common stumbles from spec §8 and early user reviews — with a one-line fix wherever possible.
Prerequisites: none — this page is a reference. Time: 3 minutes to the specific case.
Ollama is not running
ConnectionError: Could not reach Ollama at http://127.0.0.1:11434
Fix:
# Linux/macOS:
ollama serve &
# Windows:
# 1. Start menu → launch "Ollama"
# 2. or as a service: net start ollama
Verify: curl http://127.0.0.1:11434/api/version returns JSON.
Model pull failed
Error: model 'qwen3:8b' not found locally
Fix:
ollama pull qwen3:8b
# ~4.7 GB download — about 5 min on 50 Mbit/s
# Check disk space:
# Linux/macOS: df -h /
# Windows: wmic logicaldisk get size,freespace,caption
Port conflict
OSError: [Errno 98] Address already in use — port 8741
Fix:
# Who owns the port?
# Linux/macOS: lsof -i :8741
# Windows: netstat -ano | findstr :8741
# Alternative: use a different port
export COGNITHOR_API_PORT=9000
cognithor
# or
cognithor --api-port 9000
GuardrailFailure
cognithor.crew.errors.GuardrailFailure: Guardrail failed after 3 attempt(s): too long
Causes:
- The agent consistently produces output that violates the guardrail (too long / PII / not schema-conformant).
- The guardrail feedback is too vague — the agent doesn't understand what to change.
Fix:
- Sharpen the feedback: instead of "too long" →
f"Output has {count} words, max {max} allowed". - Increase
max_retries: default 2, with complex guardrails maybe 5. - Unit-test the guardrail: write a test that only calls the guardrail function — otherwise you debug blind.
Details: docs/guardrails.md.
Unknown tool
cognithor.crew.errors.ToolNotFoundError: Tool 'web_seach' not found. Did you mean 'web_search'?
Fix:
# List all available tools
cognithor tools list
# Or via Python:
python -c "from cognithor.crew.tool_resolver import available_tool_names; \
from cognithor.mcp.tool_registry_db import ToolRegistryDB; \
print(available_tool_names(ToolRegistryDB('~/.cognithor/db/tool_registry.db')))"
The did_you_mean suggestion in the error often reveals the typo directly.
Template name collision
Error: Directory 'my-project' already exists and is not empty
Fix:
# Pick a different name
cognithor init my-project-v2 --template research
# Or overwrite
cognithor init my-project --template research --force
Careful: --force does not delete — it merges. Your own changes can be overwritten — starting in an empty directory is safer.
Planner model too slow
TimeoutError: Planner took longer than 60s
Fix:
- Use a smaller model:
COGNITHOR_MODEL_PLANNER=qwen3:8binstead ofqwen3:32b. - Check GPU support:
ollama psshows load type (CPU vs GPU). cognithor --liteboots with fewer background services.
"cognithor_home not initialized"
RuntimeWarning: cognithor config load failed (...); using temp-dir tool registry
Fix:
# One-time bootstrap initialization
cognithor --init-only
# Or: manually nuke if corrupted
rm -rf ~/.cognithor/.cognithor_initialized
cognithor --init-only
Ollama installed as Windows service, ollama CLI not in PATH
'ollama' is not recognized as an internal or external command
Fix: open a new shell (PATH update requires a fresh session). If still missing: where.exe ollama — if empty, reinstall.
More help
docs/quickstart/EXTERNAL_REVIEW_RESULTS.md— currently known stumbles from external reviews.- GitHub Issues — search the error message, often already resolved.
End of Quickstart. Back to the overview.