Radar / AI infrastructure / Headroom

Headroom: the open-source context-compression layer cutting agent LLM costs 60-95%

Headroom is an Apache 2.0 context-compression layer from Headroom Labs (built by Netflix senior engineer Tejas Chopra) that sits between an AI agent and the LLM provider, compressing tool outputs, logs, RAG chunks, code and files before they reach the model. It claims ~20% fewer tokens for coding agents and 60-95% for JSON-heavy workflows with no accuracy loss on standard evals (GSM8K, TruthfulQA), using a fine-tuned compressor model (kompress-v2-base) plus reversible CCR caching so the model can retrieve originals on demand. Ships as a Python/TypeScript library, a drop-in HTTP proxy (zero code changes), a one-command agent wrap for Claude Code/Codex/Cursor/Aider/Copilot, and an MCP server.

trendinginfra · tokenizerJEV traction 0.27headroomlabs-ai/headroom · 77,708 ★added 2026-10-20

Open Headroom →View on the radar

Why it matters

The only open-source compression layer that intercepts everything an agent reads (not just conversation history), and it is reversible by design: originals are cached locally and the LLM gets a retrieval tool, so compression fails open instead of silently losing information.

What you could build with it

A small DevOps consultancy could route its client-facing coding agents through Headroom's proxy mode, cutting each client's monthly LLM bill by more than half at identical answer quality - and present the savings as a line item in its retainer.

Does it hold up?

Usable today: independent reviews confirm the savings on JSON/log-heavy agent workloads and the proxy installs with zero code changes; but dense prose compresses only 5-15%, and heavy use needs ~1-5GB of local CCR cache plus tolerance for harder-to-inspect compressed prompts.

Built with Headroom

Headroom Review: 60-95% LLM Token Compression (2026)andrew.ooo · Independent review with measured numbers (92% cut on code search, SRE incident debugging 65K to 5K tokens), documents the three deployment modes, the reversible CCR design, and corrections from Aug 2026 (repo moved to headroomlabs-ai/headroom, coding-agent claims revised down).
RAG Context Compression: Headroom Cuts Retrieval Token Costs by 57% Without Accuracy Lossdev.to · Hands-on engineering writeup showing the three integration patterns (library, proxy, MCP) with code, benchmark tables showing zero accuracy degradation, and honest notes on when compression does not help (dense prose, already-compressed data).
Headroom Explained: Open-Source Context Compression for AI Agents (60-95% Fewer Tokens)mer.vin · Walkthrough of the compression pipeline internals (CacheAligner, ContentRouter, SmartCrusher JSON, CodeCompressor AST) and how the four entry points share one lifecycle with fail-open transforms.

Learn more

in-depth review →

First spotted on github: source.

More AI infrastructure

DeepGEMM: DeepSeek's open GPU BLAS kernel libraryDeepSeek's DeepGEMM, a clean and efficient open-source BLAS kernel library for NVIDIA GPUs (up to 1550 TFLOPS on…infra · JEV 0.77OpenViking — self-evolving context database for AI agentsByteDance Volcengine's open-source (AGPL-3.0) context database that unifies agent memory, knowledge RAG, and skills…infra · JEV 0.76Modular open-sources MAX inference server, Mojo stdlib and accelerator kernelsThe core of Modular's unified AI deployment stack is now open: the MAX inference server (OpenAI-compatible endpoints)…infra · JEV 0.76ai-memory - git-backed long-term memory shared across coding agentsRust server and CLI that captures agent sessions via lifecycle hooks, compiles them into a git-backed markdown wiki…infra · JEV 0.76OmniRouteFree MIT AI gateway: one endpoint, 290+ providers and 500+ models with quota-aware auto-fallback.infra · JEV 0.75 Context Mode: context-window optimization for AI coding agentsMCP server for AI coding agents that sandboxes verbose tool output and persists session memory, cutting context use by…infra · JEV 0.74

Get the week's best AI launches, plus 3 ideas worth building

One email every Saturday. Ranked by traction, not hype. Free.