-
Notifications
You must be signed in to change notification settings - Fork 1
Troubleshooting
| Symptom | Cause | Fix |
|---|---|---|
ModuleNotFoundError |
virtualenv not active | activate .venv, reinstall requirements |
no such table |
migration not run | cd backend && python migrate.py up |
settings.json not found |
config step skipped | cp backend/settings.example.json backend/settings.json |
| Port 8766 in use | previous instance running | change --port, or stop the old process |
UnicodeEncodeError on Windows |
console is cp1252 | set PYTHONIOENCODING=utf-8
|
curl http://127.0.0.1:8766/api/system/healthThe response lists all 14 checks by name with a status each. Read it before changing anything — and note that YELLOW is not a failure:
| Verdict | Meaning | Action |
|---|---|---|
GREEN |
every subsystem reachable | none |
YELLOW |
up and running; something optional is absent | read the message — it names the remedy |
RED |
the House itself is faulty | fix the named check; ok is false
|
Two YELLOWs are normal on a fresh install and resolve themselves:
-
ollama — no local model runtime at …— you have not started a model runtime yet, or you intend to use a cloud connection instead. Both are supported states. -
SELF.md — not yet generated— the backend writes this file for itself at boot; it appears the first time the server starts.
A RED check is different in kind: a missing UI file, a module that will not import, a corrupt
database, or a broken audit chain. Those are the only conditions that make ok false.
| Symptom | Cause | Fix |
|---|---|---|
| Calls hang or time out | Ollama not running |
ollama serve, then ollama list
|
model not found |
tag not pulled |
ollama pull <tag>, and check settings.json matches |
| Tool calls produce malformed JSON | model too small | use ≥7B, or route execution to a code-tuned model |
| Long tasks fail oddly mid-run | context ceiling exceeded | raise --ctx-size on the server, or use the compact prompt |
| Cloud provider 401 | key missing or wrong api_type
|
check .env and /api/connections/{id}/ping
|
A small model failing at execution is the system working — the governance gate is rejecting malformed output rather than accepting nonsense. See Choosing a Model.
| Symptom | Cause | Fix |
|---|---|---|
| Backend cannot reach Ollama | used localhost inside the container |
it is http://ollama:11434
|
| Memory gone after restart |
down -v deleted the volume |
see Docker |
| Healthcheck never turns healthy | backend failed to boot | docker compose logs skynetclaw |
| Symptom | Likely cause |
|---|---|
| No memory of prior sessions | fresh database — the House has no history yet (First Run) |
| Verdicts keep getting rejected | Governance Engine is doing its job — the forecast lacks an invalidation condition, or a claim lacks evidence |
| A tool call is refused | the PRE_ACT gate denied it — check the governance log (Security Model) |
| Recall returns nothing useful | too few graded sessions; recall weights justified memories (Recall Quality) |
cd backend && python -m pytest -q # expect 601 passedA failure here on a clean checkout is a bug worth reporting — please include the OS, Python version, and the failing output.
→ Next: FAQ
SkynetClaw · THE HOUSE · Apache-2.0 · a council that forgets every meeting is not a council
Start
How it works
- Architecture
- The Council
- Institutional Memory
- Recall Quality
- House Mind
- Governance Engine
- Reputation
- Outcome Tracking
Using it
Running it