Multi-Backend
Switch between Copilot CLI (GitHub), Claude CLI (Anthropic), Vercel AI SDK, and Mock LLM for testing. Same interface everywhere.
@witqq/agent-sdk
One SDK, four AI backends. Type-safe tools, streaming events, and React UI components.
Type-Safe Tools
Define tools with Zod schemas. Permission scoping (once, session, project, always). Supervisor hooks for approval workflows.
Streaming Events
15 typed event types via AsyncIterable. Text deltas, thinking, tool calls, permissions, usage tracking. Custom middleware pipeline.
Testing Built In
Mock LLM backend with deterministic response modes. Mock factories for services, runtimes, and React components. No API keys needed.
Quick Install
Section titled “Quick Install”npm install @witqq/agent-sdkBackend Comparison
Section titled “Backend Comparison”| Feature | Copilot CLI | Claude CLI | Vercel AI | Mock LLM |
|---|---|---|---|---|
| Auth | GitHub OAuth | Anthropic OAuth | API key | None |
| Streaming | Yes | Yes | Yes | Yes |
| Tool calls | Yes | Yes | Yes | Simulated |
| Structured output | Yes | Yes | Yes | Configurable |
| Requires CLI | Yes | Yes | No | No |
| Cost | Copilot subscription | Pay-per-use | Pay-per-use | Free |