Skip to content

DEPLOY.md

Latest commit

 

History

History
87 lines (58 loc) · 2.19 KB

DEPLOY.md

File metadata and controls

87 lines (58 loc) · 2.19 KB
Raw

Deploying consequencegraph to Render (free tier)

What gets deployed

A read-only visual demo of the neural-lam knowledge graph. The graph is pre-built locally and committed to the repo — no indexing happens at runtime, so the free tier has plenty of CPU headroom.

Step 1 — Pre-build the graph locally

From your neural-lam root:

python consequencegraph/cli.py index ./neural_lam --preset neural_lam

This writes .consequencegraph/cache.json. Now copy it into the consequencegraph repo:

mkdir -p consequencegraph/neural_lam_cached
cp .consequencegraph/cache.json consequencegraph/neural_lam_cached/cache.json

The server in production mode loads from this cache on startup. No network calls, no indexing, instant boot.

Step 2 — Update server.py boot path

In server.py, the main() function defaults --path to ./neural_lam. For deployment, point it at the cached directory:

parser.add_argument("--path", default="./neural_lam_cached")

Or just set it in render.yaml startCommand (already done).

Step 3 — Push to GitHub

git add consequencegraph/
git commit -m "add consequencegraph with pre-built neural-lam graph"
git push

Make sure neural_lam_cached/cache.json is NOT in .gitignore. It's ~2MB, fine to commit.

Step 4 — Deploy on Render

  1. Go to https://render.com and sign up (free, no credit card)
  2. New → Web Service → Connect your GitHub repo
  3. Render will detect render.yaml automatically
  4. Deploy

Your URL will be: https://consequencegraph.onrender.com (or similar)

Production behaviour

  • /api/reindex is disabled (returns 403)
  • Rate limit: 60 requests/minute per IP
  • CORS open (anyone can embed the API)
  • Free tier sleeps after 15min inactivity — first load after sleep takes ~30s

To avoid sleep on the free tier, use UptimeRobot (free) to ping /api/stats every 14 minutes.

Local development

# Full mode — indexes live codebase, reindex enabled
python consequencegraph/server.py --path ./neural_lam --preset neural_lam --reindex

# Production simulation locally
CONSEQUENCEGRAPH_ENV=production python consequencegraph/server.py --path ./neural_lam_cached