ProMem

Your organisation memory that automates your work.

ProMem is a multi-user FastAPI app over Supabase Postgres, with Google OAuth via Supabase Auth. Reads your local productivity-tracker SQLite, classifies activities into Super-Contexts, links them to project deliverables, and synthesizes per-project wiki pages. No embeddings, no vector store, no GPU.

Quick start (local dev)

python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
python3 -m uvicorn promem_app:app --port 8888 --reload

Then open http://localhost:8888/wiki — you'll be redirected to /login for Google sign-in via Supabase.

Setup checklist (one-time)

1. Supabase project — Dashboard → New project → Mumbai (or closest).

2. Schema — paste migrations/0002_postgres_init.sql into Supabase SQL Editor → Run. Confirms 10 tables under public schema.

3. Google OAuth — enable Google provider in Supabase Auth, add the Supabase callback URL (https://<project>.supabase.co/auth/v1/callback) to Google Cloud OAuth client → Authorized redirect URIs.

4. Whitelist — add http://localhost:8888/login (and your prod URL) to Supabase Auth → URL Configuration → Redirect URLs.

5. .env — copy these from Supabase Dashboard. Do not put # comments on the same line as a value — the env loader treats # foo after an unquoted value as part of the value:

   PROMEM_DB_URL="postgresql://postgres:PASSWORD@db.PROJECT.supabase.co:5432/postgres"
   SUPABASE_URL="https://PROJECT.supabase.co"
   SUPABASE_ANON_KEY="eyJhbGciOi..."
   SUPABASE_JWT_SECRET="your-jwt-secret"
   OPENAI_API_KEY="sk-..."
   PROMEM_USER_ID="00000000-0000-0000-0000-000000000000"
   PROMEM_TRACKER_DB="/Users/you/.productivity-tracker/tracker.db"

   PROMEM_USER_ID: Supabase → Authentication → Users → click your user → "User UID" (a UUID). OPENAI_API_KEY: needed only for the nightly classify/synthesize phases. PROMEM_TRACKER_DB: optional — defaults to ${PROMEM_DATA_DIR}/tracker.db.

Running the pipeline

The nightly pipeline is local-first: it reads your local tracker.db and writes per-user data to Supabase Postgres.

python3 promem_orchestrator.py whoami        # confirm user resolution
python3 promem_orchestrator.py check-key     # confirm env wiring
python3 promem_orchestrator.py run           # full sync → classify → match → synth
python3 promem_orchestrator.py status        # last_*_at timestamps

After a run, refresh /wiki in the browser — your SC cards fill in, deliverable wikis populate.

Deploy to Fly.io

1. Install flyctl (one-time): brew install flyctl && fly auth signup (or fly auth login).
2. Launch the app from this repo:

   fly launch --no-deploy --copy-config

   Accept the existing fly.toml when prompted; pick a unique app name (e.g. promem-rohit); skip Postgres + Redis (we use Supabase).
3. Set production secrets (these are the same values from your local .env, minus the local-only ones):

   fly secrets set \
     PROMEM_DB_URL="postgresql://postgres:PASSWORD@db.PROJECT.supabase.co:5432/postgres" \
     SUPABASE_URL="https://PROJECT.supabase.co" \
     SUPABASE_ANON_KEY="eyJ..." \
     SUPABASE_JWT_SECRET="your-jwt-secret"

   Don't set PROMEM_USER_ID, PROMEM_TRACKER_DB, or OPENAI_API_KEY in Fly — those are for the local pipeline run only.
4. Whitelist the prod URL in Supabase Dashboard → Authentication → URL Configuration → Redirect URLs: https://YOUR-APP.fly.dev/login
5. Deploy: fly deploy. Wait ~2 min. Open https://YOUR-APP.fly.dev/wiki → redirects to /login → Google sign-in → wiki.

The pipeline still runs on your local machine (it needs read access to your local tracker.db); it writes to the same Supabase Postgres so the Fly-deployed UI sees the same data.

Layout

promem_app.py              FastAPI app + routes (/wiki, /projects, /productivity, /login, /auth/*)
promem_orchestrator.py     CLI driver for the nightly pipeline
promem_pipeline/
  ├── sync.py              Stage 2: tracker.db → work_pages
  ├── classify.py          Stage 3: LLM classifies into (sc_label, ctx_label)
  ├── filter.py            Stage 3.5: archive non-keep SCs
  ├── matcher.py           Stage 4: match work_pages to deliverables (5-layer scoring)
  └── synthesis.py         Stage 5: LLM synthesizes per-SC and per-deliverable wiki prose
db.py                      Postgres connection pool + user_id helper
auth.py                    Supabase JWT verification (ES256 via JWKS, HS256 fallback)
templates/                 Jinja2 templates (promem_*.html, all extend promem_base.html)
migrations/0002_*.sql      Postgres schema (multi-user with RLS)

Environment variables

Var	Required for	Purpose
PROMEM_DB_URL	app + pipeline	Supabase Postgres connection string
SUPABASE_URL	app (auth)	Project URL — used for JWKS endpoint + login page
SUPABASE_ANON_KEY	app (login page)	Public anon key, embedded in /login HTML
SUPABASE_JWT_SECRET	app (HS256 only)	Legacy projects on HS256; ES256 uses JWKS automatically
PROMEM_USER_ID	pipeline	UUID of the user whose data the pipeline writes
PROMEM_TRACKER_DB	pipeline	Path to local productivity-tracker SQLite
PROMEM_DATA_DIR	optional	Used as fallback for tracker.db lookup
OPENAI_API_KEY	pipeline	gpt-4o-mini via httpx for classify + synthesize
PROMEM_SECURE_COOKIES	optional	true in HTTPS production to enable Secure flag

License

Apache 2.0 — see LICENSE.