Headroom 0.44.0 released
A feature release adding shadow mode routing, lossless compression, per-request cost tracking, and support for additional AI agents and providers.
What changed in 0.44.0
- PyTorch extras removed from all packaging configurations.
- Shadow mode (HEADROOM_MODEL_ROUTER_LOG_ONLY) added for model routing; lossless /v1/compress mode and automatic provider detection by wire shape now available.
- Per-request cost_usd exposed in /stats; per-request metadata shown to authenticated callers; turn hooks now receive one conversation id per client conversation.
- Hermes agent and Factory Droid (headroom wrap droid) wrap support added; Serena code-memory MCP made opt-in.
- Numerous fixes including response buffering on client disconnect, Bedrock credential refresh, cache TTL handling, and tokenizer alias resolution.
Check before you update
- PyTorch extras removed from packaging
Summarised by AgentGid from Headroom Labs's official release notes on GitHub, as reported there; read the original before relying on it. Price, status and release dates come from our daily data.
Headroom at a glance
| What it is | Compresses tool outputs, logs, files, RAG chunks and conversation history before they reach the model, so agents use fewer tokens. Runs locally as a library, a proxy, an agent wrapper or an MCP server; originals stay ca… |
|---|---|
| Latest version | 0.44.0 is the latest stable release we track. |
| Release pace | 5 stable releases in the last 30 days; 0.44.0 came 5 days after the previous one |
| Price | Free (OSS). Free and open source (Apache-2.0). Compression runs on your machine. Official pricing ↗ (checked Oct 9, 2026) |
| Downloads, last 7 days | 143K down 10.2% |
| Gid Score | 54 #8 in Agent tooling for cost, context and memory |
