sarmalink-setup

name	sarmalink-setup
description	Use when setting up SarmaLink-AI for the first time, deploying to Vercel, configuring Supabase, adding API keys, or troubleshooting a fresh installation. Also use when someone says 'help me set up', 'deploy this', 'configure environment', or 'I just cloned this repo'.

SarmaLink-AI Setup Skill

You are helping a user set up SarmaLink-AI — an open-source multi-provider AI assistant with automatic failover across 36 engines and 7 providers.

Before you start

Read these files for full context:

• CLAUDE.md — project overview and architecture
• .env.example — all environment variables
• supabase/migrations/001_sarmalink_ai.sql — database schema

Setup flow — walk through each step, wait for confirmation

Step 1: Install dependencies

npm install

Verify it completes without errors. If node_modules/ already exists, skip.

Step 2: Create Supabase project

Tell the user:

1. Go to https://supabase.com and sign up (free, no credit card)
2. Click "New Project" — pick any name and a strong database password
3. Wait for the project to finish provisioning (~30 seconds)
4. Go to Settings > API and copy these three values:
   • Project URL (starts with https://)
   • anon public key (starts with eyJ)
   • service_role key (starts with eyJ, keep this secret)

Ask the user to paste all three values.

Step 3: Run database migration

Tell the user:

1. In their Supabase dashboard, go to SQL Editor
2. Click "New Query"
3. Read supabase/migrations/001_sarmalink_ai.sql and tell the user to paste it into the SQL editor
4. Click "Run" — should show "Success. No rows returned."
5. Verify: go to Table Editor — they should see ai_chat_sessions, ai_chat_usage, ai_events, ai_user_memories

Step 4: Get AI provider keys (minimum: Groq)

Required — Groq (free):

1. Go to https://console.groq.com
2. Sign in with Google
3. Create an API key
4. Copy the key (starts with gsk_)

Ask the user to paste it.

Then offer optional providers — all free, all increase capacity:

Provider	What it adds	Signup URL
SambaNova	DeepSeek V3.2 (685B) — frontier reasoning	https://cloud.sambanova.ai
Cerebras	2,000 tok/sec ultra-fast inference	https://cloud.cerebras.ai
Google Gemini	Live web search grounding	https://aistudio.google.com/app/apikey
OpenRouter	17+ free fallback models	https://openrouter.ai
Tavily	Web search (1,000/month)	https://app.tavily.com
Cloudflare	Image generation + R2 file storage	https://dash.cloudflare.com

For each one the user wants, walk them through signup and key creation. Don't push — more keys = more capacity, but Groq alone works fine.

Step 5: Create .env.local

Read .env.example and create .env.local with the user's actual values:

cp .env.example .env.local

Then edit .env.local — fill in all the keys the user provided. Ask them:

• "What do you want your assistant called?" → set NEXT_PUBLIC_APP_NAME
• "What's your company/org name?" → set NEXT_PUBLIC_COMPANY_NAME
• Set NEXT_PUBLIC_HOME_REDIRECT="/login" (or their preferred landing page)

Step 6: Verify everything works

Run in order:

npx tsc --noEmit      # Should complete with no output (= no errors)
npm test              # Should show 90 tests passing
npm run build         # Should complete with ✓ marks

If anything fails, read the error and fix it. Common issues:

• Missing env vars → check .env.local
• Module not found → run npm install again
• Supabase connection error → verify URL and keys

Step 7: Launch locally

npm run dev

Tell the user to open the URL shown (usually http://localhost:3000). They should see the app. First signup creates their account.

Step 8: Optional — Deploy to Vercel

If the user wants a public URL:

1. Push the repo to their GitHub (if not already)
2. Go to https://vercel.com → "Add New Project"
3. Import their GitHub repo
4. In "Environment Variables", add every variable from .env.local
5. Click "Deploy"
6. Wait ~60 seconds — they'll get a URL like their-app.vercel.app

Step 9: Optional — Lock down registration

If the user wants to restrict signups:

1. Supabase Dashboard → Authentication → Settings → uncheck "Enable sign-ups" to block new registrations
2. Invite-only: use Supabase's "Invite user" button to add specific emails
3. Admin endpoint: set ADMIN_EMAILS=[e-mail] in env vars to protect /api/admin/health

Step 10: Contribute back

If you fixed any bugs or improved anything during setup:

1. Create a commit describing what you changed
2. Tell the user to fork https://github.com/sarmakska/sarmalink-ai and submit a PR
3. The SarmaLink-AI project welcomes contributions — see CONTRIBUTING.md

Common mistakes

• Forgetting the database migration — the app will boot but auth and chat fail silently
• Wrong Supabase key — the anon key is the PUBLIC one, the service role key is the SECRET one. Don't swap them.
• Skipping npm install — build and tests will fail on missing modules
• Not setting NEXT_PUBLIC_APP_URL — OpenRouter requests may be rejected (wrong referer header)
• Using expired API keys — run npm run dev and try a chat message to verify each provider works

Architecture reference

User sends message
    |
Auto-router (regex, instant) --> Smart / Reasoner / Live / Fast / Coder / Vision
    |
Live tools fire first (exchange rates, weather, tracking)
    |
Failover engine (up to 14 steps):
    Step 1: SambaNova DeepSeek V3.2 --> try all keys
    Step 2: Groq GPT-OSS 120B --> try all keys
    ...
    Step 14: OpenRouter Nemotron Nano :free
    |
Each step: if 429/5xx --> next step in <50ms
    |
Stream tokens via SSE --> render markdown --> save to Supabase
    |
Extract memories in background

Built by Sarma Linux — https://sarmalinux.com