Auto — Open-Source AI Tool Hunter

Auto does not answer your question. It tells you which existing AI product to use.

Unlike generic chatbots or self-hosted agents, Auto hunts down the best ready-made tool for your task from a curated, community-editable catalog.

• Need live web research with citations? → Perplexity
• Need a landing page today? → Lovable, v0, or Bolt
• Need to refactor a large codebase? → Cursor

Table of contents

• Features
• How it works
• Quick start
• Environment variables
• Public API
• Project structure
• Roadmap
• Contributing
• License

Features

• Hybrid routing — embeddings shortlist + LLM rank (grounded to catalog only)
• Clarification — vague or ambiguous queries get follow-up questions instead of weak picks
• Recent searches — stored locally in your browser
• Feedback loop — quick result ratings saved for recommendation tuning
• Shareable results — copy link with query preserved
• Browse + search — filter the full catalog
• Public API — see /docs in the app or openapi.yaml

How it works

flowchart LR
  Query[Your query] --> Embed[Embed query]
  Embed --> Shortlist[Top 8 from catalog]
  Shortlist --> LLM[LLM rank and explain]
  LLM --> Result[Primary + alternatives]

Loading

1. Embeddings shortlist the most relevant tools from data/tools/.
2. LLM picks only from that shortlist and explains why (no hallucinated products).

Quick start

Prerequisites

• Node.js 20+
• pnpm 9+
• OpenAI API key (for embeddings at index time and recommendations at runtime)

Setup

pnpm install
pnpm build

# Generate embeddings
# Production quality (requires OPENAI_API_KEY):
export OPENAI_API_KEY=sk-...
pnpm index-catalog

# Dev/CI mock embeddings (committed by default; ranker still needs OPENAI_API_KEY):
pnpm index-catalog:mock

# Run the web app + API
pnpm dev

Open http://localhost:3000.

Deployment notes live in docs/deployment.md.

Environment variables

Create apps/web/.env.local (or repo root .env.local) with your OpenAI key:

cp apps/web/.env.example apps/web/.env.local
# Edit apps/web/.env.local and set OPENAI_API_KEY=sk-...

Restart the dev server after saving.

Variable	Description
OPENAI_API_KEY	Required — powers recommendations and catalog indexing
EMBEDDING_MODEL	Default: text-embedding-3-small
RANKER_MODEL	Default: gpt-4o-mini
INTENT_MODEL	Default: gpt-4o-mini; routes single vs workflow tasks
PLANNER_MODEL	Default: RANKER_MODEL; plans workflow steps
APP_MODE	Default: local; set open for hosted shared-key + credits
NEXT_PUBLIC_APP_URL	Public app URL for checkout redirects and docs examples
ALLOWED_ORIGIN	Optional CORS origin for hosted open mode
ADMIN_TOKEN	Sign-in secret for the admin area. Set it, then paste the same value into the Sign in form to reach /admin/feedback. Unset = admin login is disabled (login returns 403). Generate one with openssl rand -base64 24.
SESSION_SECRET	Signs admin and anonymous credit cookies. Falls back to a value derived from ADMIN_TOKEN if unset, but set a dedicated long random secret (openssl rand -base64 32) in production. Changing it invalidates existing sessions.
AUTH_SECRET	Required for Google sign-in — signs Auth.js JWT sessions. openssl rand -base64 32.
AUTH_GOOGLE_ID / AUTH_GOOGLE_SECRET	Optional — enable Continue with Google. From a Google Cloud OAuth client (redirect URI <app-url>/api/auth/callback/google). Unset = the button shows but sign-in won't complete. In open mode, an anon user's credits merge into their account on first sign-in.
FEEDBACK_STORE	Default: file; use webhook for deployed/serverless feedback
FEEDBACK_WEBHOOK_URL	Optional — receives feedback records in production
FEEDBACK_WEBHOOK_SECRET	Optional bearer token for the feedback webhook
SUPABASE_URL / SUPABASE_SERVICE_ROLE_KEY	Required when APP_MODE=open
FREE_CREDITS	Default: 30
UPSTASH_REDIS_REST_URL / UPSTASH_REDIS_REST_TOKEN	Optional shared rate limit backend
STRIPE_SECRET_KEY / STRIPE_PRICE_ID / STRIPE_WEBHOOK_SECRET	Optional credit top-ups
CREDITS_PER_PURCHASE	Default: 100

Public API

POST /api/v1/recommend

curl -X POST http://localhost:3000/api/v1/recommend \
  -H "Content-Type: application/json" \
  -d '{"query": "find SEC filings for Apple with sources", "limit": 2}'

Response:

{
  "query": "find SEC filings for Apple with sources",
  "primary": {
    "toolId": "perplexity",
    "confidence": "high",
    "reason": "..."
  },
  "alternatives": [{ "toolId": "you-com", "reason": "..." }],
  "workflowTip": "optional",
  "avoid": "optional"
}

Bring your own key (BYOK)

Pass your OpenAI key instead of using the server default:

curl -X POST http://localhost:3000/api/v1/recommend \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-your-key" \
  -d '{"query": "build a SaaS dashboard"}'

Other endpoints

Method	Path	Description
GET	/api/v1/tools	List catalog (?category=search&page=1)
GET	/api/v1/tools/:id	Single tool
POST	/api/v1/feedback	Save lightweight recommendation feedback
GET	/api/v1/health	Health + catalog version

Rate limit: 20 requests/minute per IP on /api/v1/recommend.

Project structure

apps/web/           Next.js UI + API routes
packages/catalog/   Zod schemas + YAML loader
packages/core/      Hybrid recommend engine
data/tools/         Curated tool YAML files
data/embeddings.json Precomputed vectors (generated)
data/feedback.jsonl Local feedback log (ignored)
scripts/            index-catalog, validate-catalog

Roadmap

• MCP server for agent integrations
• Production embedding index in CI
• Community voting on tool entries

Contributing

Contributions are welcome! See CONTRIBUTING.md to add or edit tools in data/tools/.

License

MIT © Auto contributors