Chutes Search

An AI-powered search engine built on Chutes, based on and giving full credit to the excellent Perplexica project.

Chutes Search serves as both a production search tool and a reference implementation for the Chutes Dev Community — showing how to build AI-powered applications using Chutes infrastructure, including OAuth ("Sign in with Chutes"), model inference, and sandbox-based deep research.

Search in action

What's different from Perplexica

• Chutes as the AI backend — LLM inference via Chutes API (llm.chutes.ai/v1), with automatic model fallbacks across multiple providers
• Sign in with Chutes — OAuth 2.0 PKCE authentication via Chutes IDP, so users can search using their own Chutes account for unlimited queries
• Free tier with rate limiting — Anonymous users get 3 free searches/day (IP-hashed, privacy-preserving) with global throttles
• Deep Research via Sandy — Extended research mode that spins up sandboxed environments using Sandy for agent-driven multi-step research
• Neon Postgres — Persistent storage via Neon serverless Postgres (replacing SQLite for serverless deployment)
• Vercel deployment — Runs on Vercel with output: 'standalone', designed for serverless
• Encrypted user data — Chat messages, titles, and auth tokens encrypted at rest (AES-256-GCM per-user)
• Optimization modes — Speed, Balanced, and Quality modes route to different models

Live demo

chutes-search.com — Try it out (3 free searches/day, or sign in with Chutes for unlimited)

Architecture

• Frontend: Next.js 15 (App Router)
• LLM: Chutes API (OpenAI-compatible), with fallback chains
• Search: Serper API for web search, SearxNG-compatible
• Database: Neon Postgres via @neondatabase/serverless + Drizzle ORM
• Auth: Chutes IDP (OAuth 2.0 PKCE)
• Deep Research: Sandy sandboxes with Claude Code agents
• Hosting: Vercel (serverless)

Focus modes

• All Mode — Searches the entire web for the best results
• Writing Assistant — Helps with writing tasks without web search
• Academic Search — Finds articles and papers for research
• YouTube Search — Finds relevant YouTube videos
• Reddit Search — Searches Reddit for discussions and opinions
• Deep Research — Multi-step agent-driven research using Sandy sandboxes (requires sign-in)

Getting started

Environment variables

# Required
DATABASE_URL=            # Neon Postgres connection string
SERPER_API_KEY=          # Serper.dev API key for web search

# Chutes AI (LLM inference)
CHUTES_API_KEY=          # Chutes API key (used for anonymous/free searches)
CHUTES_API_URL=          # e.g. https://llm.chutes.ai/v1
CHUTES_MODEL_NAME=       # e.g. deepseek-ai/DeepSeek-V3.2-TEE
MODEL_ROUTER_API_URL=    # optional fallback, defaults to https://model-router-ten.vercel.app/v1

# Chutes IDP (OAuth, optional)
CHUTES_IDP_CLIENT_ID=    # OAuth client ID
CHUTES_IDP_CLIENT_SECRET= # OAuth client secret
CHUTES_AUTH_SECRET=      # Encryption key for tokens/fields

# Deep Research (optional)
SANDY_BASE_URL=          # Sandy API endpoint
SANDY_API_KEY=           # Sandy API key

Local development

git clone https://github.com/chutesai/chutes-search.git
cd chutes-search
npm install
cp sample.config.toml config.toml  # Edit with your settings
npm run dev

Deploy to Vercel

The project is configured for Vercel with output: 'standalone'. Set the environment variables above in your Vercel project settings, then deploy.

The build runs drizzle-kit push automatically to sync the database schema.

Credits

This project is a fork of Perplexica by ItzCrazyKns. Perplexica is an outstanding open-source AI search engine that made this project possible. All original Perplexica contributors deserve recognition for their work.

License

Same license as the upstream Perplexica project — see LICENSE.