verify

name	verify
description	Launch SEPHIROTH locally and smoke-test the core flow (health, auth, one consultation) — this repo's project verify skill, with this machine's port/venv quirks baked in.

Launching and verifying SEPHIROTH locally

This machine runs other Docker stacks that squat the default ports — always use the ports below, not the framework defaults.

Port map (this machine)

Service	Default	Actual (this machine)
Postgres	5432	5433
Backend	8000	8000 (but always hit 127.0.0.1, not localhost — IPv4 only)
Frontend	3000	3100

These are set in .env (gitignored) and read by platform/core/config.py. If .env is missing, recreate it from .env.example and set GEMINI_API_KEY=<your key> and DATABASE_URL=...localhost:5433/....

Launch sequence

cd clinical-ai-copilot

# 1. Postgres
docker compose up -d postgres

# 2. Backend (platform/ is on PYTHONPATH, not a package — see CLAUDE.md)
PYTHONPATH=.:platform .venv/bin/uvicorn api.main:app --reload --port 8000 &
curl -s http://127.0.0.1:8000/health

# 3. Frontend
cd platform/frontend && npm run dev -- --port 3100 &

If .venv doesn't exist yet: python3.11 -m venv .venv && .venv/bin/pip install -r requirements.txt (system python3 is 3.9 — too old).

Smoke test

# Health
curl -s http://127.0.0.1:8000/health

# Register + capture token
TOKEN=$(curl -s -X POST http://127.0.0.1:8000/api/auth/register -H "Content-Type: application/json" \
  -d '{"email": "[e-mail]", "name": "Verify Bot", "password": "verifypass123"}' \
  | python3 -c "import sys, json; print(json.load(sys.stdin)['access_token'])")

# One real consultation through the full multi-agent workflow (calls the Gemini API — burns free-tier quota)
curl -s -X POST http://127.0.0.1:8000/api/agents/consult \
  -H "Content-Type: application/json" -H "Authorization: Bearer $TOKEN" \
  -d '{"query": "What A1C goal is appropriate for a nonpregnant adult with type 2 diabetes?"}' \
  | python3 -m json.tool

Confirm the response has a non-empty answer, citation_report.fabricated == [] (or explains why not), and agents_involved includes "evidence".

If something's wrong

• Backend 503 on /consult: Gemini isn't reachable — check GEMINI_API_KEY is set in .env and hasn't hit a quota limit. Error message names the model that failed health check.
• 429 / rate limited: the free-tier RPM quota (gemini_rpm_limit) was exceeded — the client retries with backoff automatically; if it persists, wait a minute or lower concurrent consultations.
• Import errors on backend start: confirm platform/ has no __init__.py at its root (it would shadow the stdlib platform module) and that PYTHONPATH=.:platform is set.