headroom/sdk/typescript/examples

headroom-ai examples

Runnable examples showing how to use the Headroom TypeScript SDK for context compression.

Prerequisites

• Node.js 18+
• A running Headroom proxy (pip install "headroom-ai[proxy]" && headroom proxy)
• OPENAI_API_KEY set in your environment (most examples use OpenAI)
• Optional: ANTHROPIC_API_KEY for Anthropic examples

Running

cd sdk/typescript
npm install
npx tsx examples/<filename>.ts

Examples

Vercel AI SDK

Example	Description
with-headroom-vercel.ts	One-liner withHeadroom(openai('gpt-4o')) — simplest integration
streaming-chat.ts	withHeadroom + streamText for real-time streaming
tool-calling-agent.ts	Multi-step agent with tools, context auto-compressed each step
structured-output.ts	Extract structured data with Output.object() from compressed context
middleware-composition.ts	Stack headroomMiddleware with other middlewares (extractReasoningMiddleware)
multi-provider.ts	Same compression across GPT-4o and GPT-4o-mini

Core SDK

Example	Description
basic-compress.ts	compress() function — compress then send to any LLM
simulation-dry-run.ts	simulate() — see what compression would do without calling the LLM
hooks-custom-compression.ts	CompressionHooks — customize with pre/post hooks and per-message biases
shared-context-multi-agent.ts	SharedContext — compressed handoff between agents (70-90% savings)
ccr-retrieve.ts	CCR — retrieve original content after compression (lossless)

Native SDK Adapters

Example	Description
openai-anthropic-adapters.ts	withHeadroom for native OpenAI and Anthropic SDKs (no Vercel AI SDK)