The Open Standard for Generative UI
OpenUI + Vercel Eve Harness
A generative-UI chat application backed by a Vercel Eve
agent. Eve keeps its native session and resumable-streaming protocol. OpenUI's
eveAdapter() from @openuidev/react-headless maps Eve's NDJSON stream to AG-UI.
Notes for coding agents
This project uses the eve framework. Before writing code, read the relevant guide in node_modules/eve/docs/.
The hand-written ChatLLM in src/eve-chat.ts is an intentional exception to the repo's fetchLLM-first convention: there is no HTTP chat endpoint here. Its send drives Eve's native session protocol (POST /eve/v1/session, resumable NDJSON event stream) and synthesizes the AG-UI SSE stream client-side from Eve session events, so fetchLLM does not apply — do not migrate it. Normal HTTP chat backends should use fetchLLM from @openuidev/react-ui instead of hand-writing a ChatLLM.
Prerequisites
- Node.js 24
- pnpm, npm, or Bun
- An API key for OpenUI Cloud
Run locally
Enter the standalone example and install its dependencies:
cd examples/agent-frameworks/vercel-eve pnpm install --ignore-workspaceCopy the example environment file and add your provider configuration:
cp .env.example .envSet
THESYS_API_KEYin.env. The Eve agent calls OpenUI Cloud Chat Completions as its model provider. Ask "What's the weather in Berlin?" to exercise the included Eve tool.Start the Next.js application and embedded Eve development server:
pnpm devOpen http://localhost:3000 and start a conversation.
How it works
Browser / OpenUI FullScreen
│
├─ POST /eve/v1/session or /eve/v1/session/:id
├─ GET /eve/v1/session/:id/stream?startIndex=N
│
▼
Eve HTTP channel ──► Eve agent ──► model + tools
│
▼
Eve session events ──► eveAdapter() ──► OpenUI renderer
src/app/page.tsxrenders OpenUI's<AgentInterface />withopenuiLibrary.src/eve-chat.tsdelivers turns through Eve's HTTP session protocol, returns the raw NDJSON stream, and persists session cursors per OpenUI thread.eveAdapter()maps that stream to AG-UI.agent/instructions/openui.tsinjects the generated OpenUI Lang prompt when an Eve session starts.agent/instructions/identity.mdis the agent's standing identity.agent/tools/get_weather.tsis an example Eve tool (Open-Meteo). Eve's built-inask_questionis disabled so clarifying questions stay in chat text.src/thread-store.tsstores thread metadata, transcripts, continuation tokens, and stream positions in browserlocalStorage.
Configuration
| Environment variable | Default | Purpose |
|---|---|---|
THESYS_API_KEY |
— | OpenUI Cloud API key. |
OPENUI_MODEL |
google/gemini-3.6-flash-free |
Cloud model id for Eve's Chat Completions call. |
Eve commands
The normal development command is pnpm dev. The package also exposes Eve directly:
pnpm eve:dev
pnpm eve:build
pnpm eve:start
For a production-style run, build both and start both. pnpm start runs the Eve server on port 4274 and Next.js together, because Eve's Next.js plugin forwards to that server in production:
pnpm build
pnpm start
Project layout
examples/agent-frameworks/vercel-eve/
|- agent/agent.ts # Eve model (Cloud Completions) and build config
|- agent/channels/eve.ts # Eve HTTP session channel
|- agent/instructions/identity.md # Standing agent identity
|- agent/instructions/openui.ts # Generated OpenUI Lang instructions
|- agent/tools/get_weather.ts # Example Eve tool
|- agent/tools/ask_question.ts # Disables Eve's built-in ask_question
|- src/library.ts # openuiLibrary re-export for `openui generate --spec`
|- src/lib/cloud-prompt.ts # Reads spec.json into generateSystemPrompt({ cloud: true, library })
|- src/app/page.tsx # OpenUI AgentInterface chat
|- src/eve-chat.ts # Eve session transport, eveAdapter, persistence
|- src/thread-store.ts # Browser thread and transcript storage
|- next.config.ts # Installs Eve through withEve()
Security
The example uses Eve's none() channel authentication for local development. Do not expose it
publicly in that form. Configure an authenticated Eve channel, restrict network access, and apply
the provider and tool permissions appropriate for your deployment.
Verify
pnpm verify