██╗  ██╗███████╗ █████╗ ██████╗ ██████╗  ██████╗  ██████╗ ███╗   ███╗
  ██║  ██║██╔════╝██╔══██╗██╔══██╗██╔══██╗██╔═══██╗██╔═══██╗████╗ ████║
  ███████║█████╗  ███████║██║  ██║██████╔╝██║   ██║██║   ██║██╔████╔██║
  ██╔══██║██╔══╝  ██╔══██║██║  ██║██╔══██╗██║   ██║██║   ██║██║╚██╔╝██║
  ██║  ██║███████╗██║  ██║██████╔╝██║  ██║╚██████╔╝╚██████╔╝██║ ╚═╝ ██║
  ╚═╝  ╚═╝╚══════╝╚═╝  ╚═╝╚═════╝ ╚═╝  ╚═╝ ╚═════╝  ╚═════╝ ╚═╝     ╚═╝
                  The context compression layer for AI agents

60–95% fewer tokens · library · proxy · MCP · 6 algorithms · local-first · reversible

Docs · Install · Proof · Agents · Discord · llms.txt · Enterprise

_{AI agents / LLMs: read /llms.txt here, or fetch the live index / full docs blob.}

---

Headroom compresses everything your AI agent reads — tool outputs, logs, RAG chunks, files, and conversation history — before it reaches the LLM. Same answers, fraction of the tokens.

_{Live: 10,144 → 1,260 tokens — same FATAL found.}

What it does

• Library — compress(messages) in Python or TypeScript, inline in any app
• Proxy — headroom proxy --port 8787, zero code changes, any language
• Agent wrap — headroom wrap claude|codex|aider|copilot|opencode in one command; Cursor prints manual proxy settings to paste into the app
• MCP server — headroom_compress, headroom_retrieve, headroom_stats for any MCP client
• Cross-agent memory — shared store across Claude, Codex, Gemini, auto-dedup
• headroom learn — mines failed sessions, writes corrections to CLAUDE.md / AGENTS.md
• Output token reduction — trims what the model writes back (not just what you send): drops ceremony/restated code and skips deep "thinking" on routine steps. See Output token reduction.
• Reversible (CCR) — originals are cached for retrieval on demand

How it works (30 seconds)

 Your agent / app
   (Claude Code, Cursor, Codex, LangChain, Agno, Strands, your own code…)
        │   prompts · tool outputs · logs · RAG results · files
        ▼
    ┌────────────────────────────────────────────────────┐
    │  Headroom   (runs locally — your data stays here)  │
    │  ────────────────────────────────────────────────  │
    │  CacheAligner  →  ContentRouter  →  CCR            │
    │                    ├─ SmartCrusher   (JSON)        │
    │                    ├─ CodeCompressor (AST)         │
    │                    └─ Kompress-base  (text, HF)    │
    │                                                    │
    │  Cross-agent memory  ·  headroom learn  ·  MCP     │
    └────────────────────────────────────────────────────┘
        │   compressed prompt  +  retrieval tool
        ▼
 LLM provider  (Anthropic · OpenAI · Bedrock · …)

• ContentRouter — detects content type, selects the right compressor
• SmartCrusher / CodeCompressor / Kompress-base — compress JSON, AST, or prose
• CacheAligner — stabilizes prefixes so provider KV caches actually hit
• CCR — stores originals locally; LLM calls headroom_retrieve if it needs them

→ Architecture · CCR reversible compression · Kompress-v2-base model card

Get started (60 seconds)

# 1 — Install
pip install "headroom-ai[all]"          # Python
npm install headroom-ai                 # Node / TypeScript

# 2 — Pick your mode
headroom wrap claude                    # wrap a coding agent
headroom proxy --port 8787              # drop-in proxy, zero code changes
# or: from headroom import compress      # inline library

# 3 — See the savings
headroom perf
headroom dashboard                      # live savings dashboard (proxy must be running)

Granular extras: [proxy], [mcp], [ml], [code], [memory], [relevance], [image], [agno], [langchain], [evals], [pytorch-mps] (Apple-GPU memory-embedder offload — set HEADROOM_EMBEDDER_RUNTIME=pytorch_mps). Requires Python 3.10+.

Proof

Savings on real agent workloads:

Workload	Before	After	Savings
Code search (100 results)	17,765	1,408	92%
SRE incident debugging	65,694	5,118	92%
GitHub issue triage	54,174	14,761	73%
Codebase exploration	78,502	41,254	47%

Accuracy preserved on standard benchmarks:

Benchmark	Category	N	Baseline	Headroom	Delta
GSM8K	Math	100	0.870	0.870	±0.000
TruthfulQA	Factual	100	0.530	0.560	+0.030
SQuAD v2	QA	100	—	97%	19% compression
BFCL	Tools	100	—	97%	32% compression

Reproduce: python -m headroom.evals suite --tier 1 · Full benchmarks & methodology

Output token reduction (cut what the model writes back)

Everything above shrinks the prompt you send. But you also pay for every token the model writes back — and on Opus-class models output costs 5× input. A lot of that output is waste: "Great, let me…" preambles, re-printing code you just showed it, and deep "thinking" on routine steps like reading a file.

Headroom can trim that too, from the proxy, without you changing any code:

• Verbosity steering — appends a short "be terse, don't restate context" note to the end of the system prompt (so your prompt cache still hits).
• Effort routing — when a turn is just the model resuming after a tool result (a file read, a passing test), it dials the model's thinking effort down. New questions and errors keep full effort.

Turn it on:

export HEADROOM_OUTPUT_SHAPER=1     # off by default
headroom proxy --port 8787

Already running a proxy? These switches are read live on every request, so a proxy that headroom wrap reused (rather than started) would not see a value you export afterwards — its environment was snapshotted at launch. headroom wrap now hot-syncs your current settings to the running proxy via a loopback POST /admin/runtime-env, so they take effect immediately with no restart (no cold start, no dropped requests, no lost caches). Set them before you wrap. On a shared proxy these overrides are global — the last explicit setting wins.

Learn the right terseness for you. People don't say how terse they want answers — they show it (they interrupt long replies, or move on before they could have read them). headroom learn --verbosity reads your past sessions and picks the level automatically:

headroom learn --verbosity            # preview what it found (dry run)
headroom learn --verbosity --apply    # save it; the proxy uses it from now on

See how many output tokens you saved. Output savings are counterfactual — we never see what the model would have written — so Headroom reports an honest estimate with a confidence range, never a made-up number:

headroom output-savings
# Reduction: 31.7%  (95% CI 27.7% … 35.7%)   [estimated]

Want a measured number instead of an estimate? Leave 10% of conversations unshaped as a control group: export HEADROOM_OUTPUT_HOLDOUT=0.1. The dashboard shows an Output Tokens Saved card next to input compression, labelled measured or estimated with the confidence band.

→ Full write-up incl. the measurement methodology: docs/proposals/output-token-reduction.md

Agent compatibility matrix

Agent	headroom wrap	Notes
Claude Code	✅	--memory · --code-graph · --1m
Codex	✅	shares memory with Claude
Cursor	Manual setup	starts proxy and prints base URLs for Cursor settings
Aider	✅	starts proxy + launches
Copilot CLI	✅	starts proxy + launches
OpenClaw	✅	installs as ContextEngine plugin
OpenCode	✅	injects config · starts proxy + launches
Cortex Code	✅	60–65% savings · library mode

Any OpenAI-compatible client works via headroom proxy. MCP-native: headroom mcp install.

GitHub Copilot CLI subscription mode

Headroom can route GitHub Copilot CLI subscription traffic through the local proxy:

headroom copilot-auth login
headroom wrap copilot --subscription -- --model gpt-4o

This lets Headroom intercept OpenAI-compatible Copilot CLI requests and apply the same proxy compression pipeline before forwarding to GitHub Copilot's hosted API. The wrapper exchanges Headroom's reusable GitHub OAuth token for Copilot's short-lived API token and prints the upstream endpoint as COPILOT_PROVIDER_API_URL=... during launch.

headroom copilot-auth login stores a Headroom-specific Copilot OAuth token. This avoids relying on generic GitHub or Copilot CLI tokens that can read Copilot account metadata but may still be rejected by Copilot's token-exchange endpoint.

For GitHub Enterprise Server or custom-domain Copilot deployments, set the deployment domain before launching:

export GITHUB_COPILOT_ENTERPRISE_DOMAIN=ghe.example.com

For GitHub.com Enterprise Cloud URLs such as github.com/enterprises/your-enterprise, do not set an enterprise-domain override. Headroom uses GitHub's normal token-exchange endpoint and the Copilot API endpoint advertised for the signed-in account.

Platform support note: macOS auth reuse via Copilot CLI Keychain storage has been smoke-tested. Windows Credential Manager, Linux Secret Service / secret-tool, and Docker/CI token-injection paths are implemented or planned as auth-discovery paths, but still need real OS validation before they should be considered fully vetted. For Docker and CI, prefer passing an explicit GITHUB_COPILOT_TOKEN or GITHUB_COPILOT_GITHUB_TOKEN rather than relying on host keychain access.

When to use · When to skip

Great fit if you…

• run AI coding agents daily and want savings without changing your code
• work across multiple agents and want shared memory
• need reversible compression — originals are retrievable via CCR within the configured TTL

Skip it if you…

• only use a single provider's native compaction and don't need cross-agent memory
• work in a sandboxed environment where local processes can't run

Integrations — drop Headroom into any stack

Your setup	Hook in with
Any Python app	compress(messages, model=…)
Any TypeScript app	await compress(messages, { model })
Anthropic / OpenAI SDK	withHeadroom(new Anthropic()) · withHeadroom(new OpenAI())
Vercel AI SDK	wrapLanguageModel({ model, middleware: headroomMiddleware() })
LiteLLM	litellm.callbacks = [HeadroomCallback()]
LangChain	HeadroomChatModel(your_llm)
Agno	HeadroomAgnoModel(your_model)
Strands	Strands guide
ASGI apps	app.add_middleware(CompressionMiddleware)
Multi-agent	SharedContext().put / .get
MCP clients	headroom mcp install

What's inside

• SmartCrusher — universal JSON: arrays of dicts, nested objects, mixed types.
• CodeCompressor — AST-aware for Python, JS, Go, Rust, Java, C++.
• Kompress-base — our HuggingFace model, trained on agentic traces.
• Image compression — 40–90% reduction via trained ML router.
• CacheAligner — stabilizes prefixes so Anthropic/OpenAI KV caches actually hit.
• IntelligentContext — score-based context fitting with learned importance.
• CCR — reversible compression; LLM retrieves originals on demand.
• Cross-agent memory — shared store, agent provenance, auto-dedup.
• SharedContext — compressed context passing across multi-agent workflows.
• headroom learn — plugin-based failure mining for Claude, Codex, Gemini.

Pipeline internals

Headroom exposes one stable request lifecycle across compress(), the SDK, and the proxy:

Setup → Pre-Start → Post-Start → Input Received → Input Cached → Input Routed → Input Compressed → Input Remembered → Pre-Send → Post-Send → Response Received

• Transforms do the work: CacheAligner, ContentRouter, SmartCrusher, CodeCompressor, Kompress-base, IntelligentContext / RollingWindow.
• Pipeline extensions observe or customize lifecycle stages via on_pipeline_event(...).
• Compression hooks sit alongside the canonical lifecycle as an additional extension seam.
• Proxy extensions remain the server/app integration seam for ASGI middleware, routes, and startup policy.

Provider and tool-specific behavior lives under headroom/providers/ so core orchestration stays focused on lifecycle, sequencing, and policy.

• CLI/tool slices: headroom/providers/claude, copilot, codex, openclaw
• Provider runtime slices: headroom/providers/claude, gemini, plus shared backend/runtime dispatch in headroom/providers/registry.py
• Core files stay orchestration-first: wrap.py, client.py, cli/proxy.py, and proxy/server.py delegate provider-specific env shaping, API target normalization, backend selection, and transport dispatch.

Install

pip install "headroom-ai[all]"          # Python, everything
npm install headroom-ai                 # TypeScript / Node
docker pull ghcr.io/chopratejas/headroom:latest

Granular extras: [proxy], [mcp], [ml] (Kompress-base), [code], [memory], [relevance], [image], [agno], [langchain], [evals], [pytorch-mps] (Apple-GPU memory-embedder offload — set HEADROOM_EMBEDDER_RUNTIME=pytorch_mps). Requires Python 3.10+.

Using pipx? Choose a supported interpreter explicitly:

pipx install --python python3.13 "headroom-ai[all]"

Pick 3.13 if you want dollar savings. The dashboard's Proxy $ Saved tile prices compression with LiteLLM, and LiteLLM can't be installed on Python 3.14+. On 3.14 token savings still track, but the dollar figure stays $0.00. If you already installed on 3.14, switch with pipx reinstall headroom-ai --python python3.13 and restart the proxy.

→ Installation guide — Docker tags, persistent service, PowerShell, devcontainers.

Updating

headroom update          # detects pip / pipx / uv tool and upgrades in place
headroom update --check  # report the latest release without upgrading
headroom update --pre    # include pre-releases

headroom update figures out how Headroom was installed (pip/venv, pip --user, pipx, uv tool) and runs the matching upgrade across macOS, Linux, and Windows. For git checkouts, editable installs, Docker images, and externally-managed system Pythons (PEP 668) it prints the correct manual step instead of guessing.

The proxy also shows a one-line "update available" notice on startup. It checks PyPI at most once a day, in the background, and never blocks. Opt out with HEADROOM_UPDATE_CHECK=off (also skipped in --stateless mode and CI).

Corporate / SSL-inspection environments

If pip install "headroom-ai[all]" fails with CERTIFICATE_VERIFY_FAILED (unable to get local issuer certificate), your network uses SSL inspection — a MITM proxy presenting a company-issued CA. The build backend (maturin) downloads rustup over a connection your TLS stack doesn't trust. Install Rust first so the build doesn't fetch it:

# macOS / Linux
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh && rustup default stable
# Windows
winget install Rustlang.Rustup && rustup default stable

Restart your shell, then pip install "headroom-ai[all]". A prebuilt wheel avoids the Rust build entirely where available: pip install --only-binary headroom-ai headroom-ai. Prebuilt wheels are published for Windows (win_amd64), Linux (x86_64 / aarch64), and macOS (Apple Silicon), so installs on those platforms never need a local Rust toolchain — the Rust-first dance above is only for the platform-independent sdist fallback (e.g. Intel macOS).

Two runtime assets are fetched over TLS; if they are blocked, trust your corporate CA via REQUESTS_CA_BUNDLE / SSL_CERT_FILE / CURL_CA_BUNDLE:

• cdn.pyke.io — the ONNX Runtime for the Rust core. Alternatively pre-provide it with ORT_STRATEGY=system and ORT_LIB_LOCATION=/path/to/onnxruntime.
• huggingface.co — the kompress-base compression model. Pre-download it and run with HF_HUB_OFFLINE=1, or set HF_ENDPOINT to a trusted mirror.

Running with compression disabled (pure gateway) requires neither asset.

"Basic Constraints of CA cert not marked critical" (Python 3.13+ strict mode)

A different failure from the one above. If TLS fails with:

[SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed:
Basic Constraints of CA cert not marked critical

then the corporate CA is found and trusted — adding it to a CA bundle changes nothing. Python 3.13 + OpenSSL 3.x enable VERIFY_X509_STRICT by default, which enforces RFC 5280 §4.2.1.9: a CA cert's basicConstraints must be marked critical. Inspection roots like Zscaler set CA:TRUE without the critical bit, so the chain is rejected.

Set HEADROOM_TLS_STRICT=0 to clear only the strict flag from every TLS context Headroom controls — the proxy's httpx upstream client and the urllib3/huggingface_hub path used for model downloads. Chain validation, signature, expiry, and hostname checks all stay on; this is strictly narrower than disabling verification.

HEADROOM_TLS_STRICT=0 headroom proxy --port 8787

The Rust core's ONNX download (cdn.pyke.io) uses a separate TLS stack (rustls / OS trust store), unaffected by HEADROOM_TLS_STRICT. On Windows the corporate root must be in the machine certificate store (browsers already trust it there); or pre-provision ONNX Runtime with ORT_STRATEGY=system + ORT_LIB_LOCATION=/path/to/onnxruntime to skip the download entirely.

headroom learn

headroom learn — mines failed sessions, writes corrections to CLAUDE.md / AGENTS.md / GEMINI.md.

Documentation

Start here	Go deeper
Quickstart	Architecture
Proxy	How compression works
MCP tools	CCR — reversible compression
Memory	Cache optimization
Failure learning	Benchmarks
Configuration	Limitations

Compared to

Headroom runs locally, covers every content type, works with every major framework, and is reversible.

	Scope	Deploy	Local	Reversible
Headroom	All context — tools, RAG, logs, files, history	Proxy · library · middleware · MCP	Yes	Yes
RTK	CLI command outputs	CLI wrapper	Yes	No
lean-ctx	CLI commands, MCP tools, editor rules	CLI wrapper · MCP	Yes	No
Compresr, Token Co.	Text sent to their API	Hosted API call	No	No
OpenAI Compaction	Conversation history	Provider-native	No	No

Attribution. Headroom ships with the excellent RTK binary for shell-output rewriting — git show --short, scoped ls, summarized installers. Huge thanks to the RTK team; their tool is a first-class part of our stack, and Headroom compresses everything downstream of it. Headroom can also use lean-ctx as the selected CLI context tool; set HEADROOM_CONTEXT_TOOL=lean-ctx before running headroom wrap ....

Contributing

git clone https://github.com/chopratejas/headroom.git && cd headroom
uv sync --extra dev && uv run pytest

Devcontainers in .devcontainer/ (default + memory-stack with Qdrant & Neo4j). See CONTRIBUTING.md.

Community

• Discord — questions, feedback, war stories.
• Kompress-v2-base on HuggingFace — the model behind our text compression.

License

Apache 2.0 — see LICENSE.