Convex

Convex

Overview

Convex is a reactive backend platform where database queries, mutations, and actions are defined in TypeScript and data automatically syncs to connected clients in real-time. It eliminates WebSocket code, polling, and cache invalidation, providing ACID transactions and optimistic updates out of the box.

Instructions

• When defining schemas, use defineSchema() with defineTable() and typed validators (v.string(), v.number(), v.id("tableName")), and define indexes for all filtered and sorted queries.
• When writing functions, use queries for reads (automatically reactive), mutations for writes (transactional, triggers reactive updates), and actions for external API calls (non-transactional).
• When building React UIs, use useQuery() for reactive data subscriptions that auto-update, useMutation() for writes with optimistic updates, and usePaginatedQuery() for infinite scroll.
• When handling authentication, use convex-auth for built-in auth or integrate Clerk/Auth0, and validate user identity at the start of every mutation with ctx.auth.getUserIdentity().
• When processing background work, use ctx.scheduler.runAfter() for delayed execution and cron jobs for recurring tasks instead of making mutations slow.
• When storing files, use ctx.storage.store() for upload and ctx.storage.getUrl() for serving URLs without S3 or CDN configuration.
• When implementing search, use full-text search indexes with searchIndex() or vector search with vectorIndex() for AI/RAG applications, with metadata filtering.

Examples

Example 1: Build a real-time chat application

User request: "Create a real-time chat app with Convex and React"

Actions:

1. Define messages table with schema, author reference, and timestamp index
2. Create a query function that returns messages sorted by timestamp
3. Create a mutation for sending messages with auth validation
4. Use useQuery() in React to subscribe to messages with automatic real-time updates

Output: A chat application where messages appear instantly for all connected users without WebSocket code.

Example 2: Add full-text and vector search

User request: "Implement search across articles with both keyword and semantic search"

Actions:

1. Define search index on article body field with searchIndex()
2. Define vector index on embedding field with vectorIndex()
3. Create query functions for text search and vector similarity search
4. Combine metadata filtering with search for scoped results

Output: A dual search system supporting both keyword matching and semantic similarity queries.

Guidelines

• Use schema validation in production: defineSchema() catches type errors at deploy time, not runtime.
• Define indexes for all filtered/sorted queries to ensure efficient data access.
• Use queries for reads, mutations for writes, actions for external APIs; never mix concerns.
• Keep mutations small and fast since they hold a database lock; move heavy processing to actions.
• Use ctx.scheduler.runAfter() for background work instead of making mutations slow.
• Validate user identity at the start of every mutation to prevent unauthorized writes.
• Use optimistic updates for interactive UIs; the client sees the change instantly while the server confirms.