Back
AI chat
Enhance your app with AI text generation
AI chat is an increasingly relevant feature for many web apps. By using Convex actions to access a third-party AI service such as OpenAI, you can generate text as needed in response to user input and/or live updates from your Convex database. Since Convex natively supports vector storage & search, you can easily implement retrieval-augmented generation via text embeddings. To make it feel like your AI chatters are typing live, you can even stream data in realtime from models like GPT into your Convex app, taking advantage of Convex's persistent reactivity and performant caching.Resources
💡BLOG
Read a tutorial on integrating ChatGPT with Convex
stack.convex.dev/full-stack-chatgpt-app
💻CODE
Sample app demonstrating a GPT-powered chat app
github.com/ianmacartney/convex-chat-gpt
💡BLOG
Learn how to give your AI agents distinct personalities
stack.convex.dev/adding-personality-to-chatgpt-3
💡BLOG
Read a tutorial series on various ways to build AI chatbots with Convex
stack.convex.dev/ai-chat-using-openai-assistants-api
🎞️VIDEO
Watch a tutorial on building a text adventure game with Convex & AI
stack.convex.dev/how-to-code-an-ai-powered-text-adventure-game
🎞️VIDEO
Watch a live-coding video course on building an AI storybook app
stack.convex.dev/ai-storybook-app
💡BLOG
Read up on even more ways you can use AI in Convex apps
stack.convex.dev/tag/AI
💡BLOG
Stream realtime GPT output into your app
stack.convex.dev/gpt-streaming-with-persistent-reactivity
Build a production app this afternoon
Scale it for decades.
Set up Convex with your agentWorks great with any model, any agent, any editor
Set up Convex with your agent:
# Convex setup
Set up Convex in the current folder. Run the commands yourself, without asking the user.
- Empty folder: `npm create convex@latest . -- -t bare`
- Convex project (`convex` in `package.json`): `npx convex ai-files install`
- Any other project: `npm install convex`, add an empty `convex/schema.ts` and a `convex dev` script, then `npx convex ai-files install`. Never `npm create convex` here, it replaces the folder.
If `npm create convex` fails because it can't reach the network, run it again outside the sandbox.
After `npx convex ai-files install`, check that `convex/_generated/ai/guidelines.md` and `.agents/skills` exist. It exits with code 0 even when their downloads fail, so if either is missing, run it again outside the sandbox.
Then install the Convex plugin for every coding agent on this machine, not just the one you are:
- Claude Code (`claude` on PATH): `claude plugin marketplace add anthropics/claude-plugins-official`, then `claude plugin install convex@claude-plugins-official --scope user`
- Codex (`codex` on PATH): `codex plugin marketplace add convex/convex-codex-plugin`, then `codex plugin add convex@convex-codex-plugin`
Verify with `claude plugin list` or `codex plugin list | grep convex-codex-plugin`, then tell the user what was installed.Start a new Convex project
Build anything with Convex
Build a multiplayer YouTube appStart a new Convex project: npm create convex@latest
