saniplus-chatbot / DEPLOY.md
Sebacu's picture
Claude Opus 4.8
SaniPlus chatbot (LLM-backed) showcase
45a648e
|
Raw History Blame Contribute Delete
3.38 kB
# Deploying the SaniPlus chatbot to Render
A public, shareable showcase URL in ~10 minutes. The app is protected by a
shared access code, and your Anthropic API key stays server-side.
## What you need
- A GitHub account
- A Render account (https://render.com β€” free tier is fine)
- Your Anthropic API key (`sk-ant-...`)
- An access code you invent (e.g. `saniplus2026`)
---
## Step 1 β€” Put this folder in a GitHub repo
From inside `saniplus-chatbot-llm/`:
```powershell
git init
git add .
git commit -m "SaniPlus chatbot (LLM-backed) showcase"
# create an empty repo on github.com first, then:
git remote add origin https://github.com/<you>/saniplus-chatbot.git
git branch -M main
git push -u origin main
```
`.gitignore` already excludes `node_modules` and any `.env`, so no secrets are
committed.
> Push **this folder** as the repo root (so `package.json` and `render.yaml`
> are at the top level). If you push the parent folder instead, set
> `rootDir: saniplus-chatbot-llm` under the service in `render.yaml`.
## Step 2 β€” Create the service on Render
**Option A β€” Blueprint (uses the included `render.yaml`):**
1. Render dashboard β†’ **New** β†’ **Blueprint**.
2. Connect your repo. Render reads `render.yaml` and proposes the service.
3. It will ask for the two secret values (they have `sync: false`):
- `ANTHROPIC_API_KEY` = your `sk-ant-...`
- `ACCESS_CODE` = the password you chose
4. **Apply** β†’ first build runs `npm install` and starts `npm start`.
**Option B β€” Manual web service:**
1. **New** β†’ **Web Service** β†’ connect the repo.
2. Runtime **Node**, Build `npm install`, Start `npm start`.
3. **Environment** β†’ add `ANTHROPIC_API_KEY`, `ACCESS_CODE`, and
`NODE_VERSION=20`.
4. Create.
## Step 3 β€” Open it
Render gives you `https://saniplus-chatbot.onrender.com` (or similar). Open it,
enter the access code, and the chat goes live. Share the URL **and** the code
with whoever you're demoing to.
---
## Notes
- **Cold start:** the free plan sleeps after ~15 min idle; the first hit then
takes ~30s to wake. Upgrade to a paid instance ($7/mo) for always-on.
- **Health check:** Render pings `/api/health` to confirm the service is up.
- **Rotating the code:** change `ACCESS_CODE` in the Render dashboard and
redeploy β€” existing browser sessions are revalidated on next request.
- **Cost control:** the access gate keeps usage to people you share the code
with. For extra safety you can still add per-IP rate limiting (ask me).
- **Verify hours per branch** and complete the GDPR/HWG review before using
this with real customers (see README.md).
## Protection layers (both active)
- **Access code** β€” `ACCESS_CODE` gates `/api/chat` (401 without a valid code).
- **Rate limiting** β€” per-IP sliding window on `/api/chat` and `/api/gate`.
Defaults: 20 requests / 60s. Override with env vars `RATE_MAX` and
`RATE_WINDOW_MS`. Returns HTTP 429 with `Retry-After` when exceeded.
## Always-on alternative β€” Fly.io (no cold start)
The repo includes a `Dockerfile` + `fly.toml`. Deploy:
```powershell
fly launch --no-deploy # accept the included fly.toml
fly secrets set ANTHROPIC_API_KEY=sk-ant-... ACCESS_CODE=yourcode
fly deploy
```
Set `min_machines_running = 1` in `fly.toml` for guaranteed always-on (no cold
start). The same Docker image runs on Railway, Cloud Run, or any container host.