github headroomlabs-ai/headroom v0.45.0
Release v0.45.0

one hour ago

0.45.0 (2026-11-22)

Features

  • opencode: add repeatable --extra-model flag for custom wrap-time models (#2364) (66e0086)
  • savings: book modelled output savings per request, as a token and dollar range (#4878) (00436a8)
  • shaper: let an extension replace a level's steering text; default to L3 (#4877) (a3afd43)
  • telemetry: split beacon sessions by harness, model and day (#4881) (ac932eb)
  • tray: opt-in system tray icon with savings counter, dashboard and update alert (#4726) (e9301f3)

Bug Fixes

  • benchmarks: drop the duplicate protect_recent=0 table from the proof harness (#4754) (a385ea8)
  • benchmarks: refuse a dirty or non-git --worktree in codemode_ab (#4736) (6b7b635)
  • binaries: ignore tool binaries found via the current directory (#4838) (7a0f15f)
  • cache: exclude cold starts from TTL hit evidence (#4728) (93df7ae)
  • cache: preserve complete CCR hashes across restarts (#4820) (e27f0d9)
  • cache: replay every parallel tool result (#4818) (973c12d)
  • cache: scope learned TTL cache to its source file (#4727) (acb58a0)
  • ccr: send batch continuations to the upstream the batch was submitted to (#4848) (5f82080)
  • ci: keep alert rule and location out of the public CodeQL triage issue (#4863) (047aadd)
  • ci: verify installer provenance in build workflows (#4677) (b235bc2)
  • compression: bound the JSON-block and truncation-banner scans (#4845) (bae62bc)
  • compress: stop adding markers the caller did not ask for (#4678) (c21b081)
  • content-router: keep list-form tool_result text blocks on separate lines (#4739) (f4e3ad4)
  • content-router: protect cache_control markers nested in tool_result content (#4898) (18067b1)
  • copilot: keep new VS Code chats on the Local harness so they reach the proxy (#4899) (0d203db)
  • deps: move the wrap E2E harness to npm 12.2.2 and drop the fixed audit exceptions (#4778) (f6fbac3)
  • deps: patch quinn-udp, rustix and opentelemetry-api advisories (#4835) (0368a87)
  • examples: keep the macos launchagent running when reinstall is declined (#4734) (02a2b56)
  • examples: validate the macos launchagent port as plain decimal (#4733) (a91c48a)
  • gemini: apply the request rate limit to Cloud Code and countTokens (#4862) (c2292ed)
  • kompress: honour toin_read_only in the Kompress CCR store (#4895) (8da7e17)
  • learn: only fold real linked worktrees into their main checkout (#4839) (d2fb350)
  • learn: render non-string Bash commands without raising (#4826) (cfdeb1c)
  • lossless: never grep-fold Go, nginx, klog or access-log timestamps (#4901) (cf32846)
  • lossless: never grep-fold Ruby Logger timestamps (#4907) (4a94500)
  • mcp: check Host and Origin on the Streamable HTTP transport (#4852) (70ba721)
  • memory: dispatch the advertised memory_list tool (#4830) (0d91550)
  • memory: include older memories in agent sync (#4832) (a6a825c)
  • memory: keep Neo4j relationship types to a single identifier (#4837) (b0a8c00)
  • memory: return stored id from a background DirectMem0 save (#4759) (6665940)
  • memory: scope TrafficLearner's SQLite reads to its user_id (#4850) (9321a41)
  • memory: validate embedding dimensions before writes (#4831) (cd6f438)
  • oauth2: stop following redirects from the token endpoint (#4849) (02dfdd8)
  • openclaw: keep the Python launcher fallback off the working directory (#4843) (acebf20)
  • openclaw: let the circuit breaker open on proxy failures (#4755) (1372f95)
  • openclaw: preserve assistant text blocks when tool arguments change (#4829) (d077263)
  • openclaw: return structured retrieval tool results (#4707) (e6b5777)
  • opencode: default options-form https.request calls to https (#4749) (12b5fb7)
  • opencode: do not break every request when the proxy URL is empty or invalid (#4750) (318e8e0)
  • opencode: keep child process options when the args slot is null or undefined (#4748) (9f2cc01)
  • opencode: override effect to patched 4.0.0-beta.103 (#4762) (19d439f)
  • opencode: percent-encode the project name in x-headroom-project (#4780) (863b3c7)
  • opencode: stop routing a host's bare /messages or /responses to the proxy (#4781) (#4791) (87c1293)
  • opencode: stop routing non-LLM hosts and paths to the proxy (#4747) (#4758) (c48778c)
  • output-shaper: don't add Responses instructions to a body already steered in input (#4937) (3a8270b)
  • plugins: keep the hook launcher's Python fallback off the project dir (#4842) (c5d8f13)
  • proxy: bound brotli output per step and lossless inverse expansion (#4861) (5503383)
  • proxy: check x-headroom-base-url off the event loop in chat and responses (#4856) (2f58877)
  • proxy: clamp /transformations/feed limit to [1, 100] (ec2ab21)
  • proxy: clamp /transformations/feed limit to [1, 100] (#4934) (ec2ab21)
  • proxy: don't take Claude Code's "Tool loaded." note as the user query (#4714) (dcd03b9)
  • proxy: enforce HEADROOM_PROXY_TOKEN in the Rust front and listen on loopback by default (#4840) (f5c4217)
  • proxy: give the OpenAI Responses error frame its type discriminator (#4331) (53a2521)
  • proxy: keep Copilot requests on Copilot when an extension advises another provider (#4894) (998d710)
  • proxy: keep credentials out of proxy logs and 400 debug dumps (#4865) (dffc4dc)
  • proxy: keep hop-by-hop request headers off upstream calls (#4775) (8f47a86)
  • proxy: keep relay tasks when the same WS handle re-registers (#4904) (f4691e8)
  • proxy: keep reserved openai tool names resident during tool search deferral (#4753) (62f2a2d)
  • proxy: keep settings.json mode and symlink in the cc-switch reconciler (#4844) (3bae009)
  • proxy: keep upstream credentials out of manifests, env and TOIN (#4866) (2e4ff6f)
  • proxy: let an extension declare its install() fail-closed (#4737) (#4773) (c05fbdd)
  • proxy: log client-supplied model ids as one escaped token (#4854) (b9f6255)
  • proxy: make the Rust front's native Bedrock and Vertex routes opt-in (#4851) (99021af)
  • proxy: pin non-ASCII upstream hosts by A-label and resolve x-headroom-base-url once (#4855) (904e56a)
  • proxy: refuse a private x-headroom-base-url on the async chat and responses path (#4879) (5023c84)
  • proxy: refuse HTTP requests from untrusted browser origins (#4853) (ab1d029)
  • proxy: reject a private x-headroom-base-url on chat and responses with 400 (#4857) (491afe9)
  • proxy: resolve the forwarded client from the right and keep settings writes loopback-only (#4847) (e15e342)
  • proxy: retain final backend streaming usage (#4814) (93331f2)
  • proxy: retry a TLS error only while the request is being written (#4776) (854a696)
  • proxy: send operator extra headers only to a designated origin (#4858) (6d8121c)
  • proxy: stop a "*" origin setting from waiving the WebSocket Origin check (#4846) (d9ae7ad)
  • proxy: turn off the image stage when --compressor leaves out image (#4723) (8f04221)
  • router: keep source code out of the prose (Kompress) fallbacks (#4787) (3aafdf2)
  • sdk: accept string contents and functionCall without args in the Gemini adapter (#4756) (200c838)
  • sdk: encode the signature hash in client.telemetry.getTool (#4772) (3c66c91)
  • sdk: keep Gemini thought signatures and call ids through the compress round trip (#4777) (622df29)
  • sdk: reject "." ".." and NUL plugin names like the python twin (#4771) (99ef4ee)
  • sdk: reject sharedcontext maxentries below 1 (#4743) (4b6529f)
  • sdk: throw typed proxy 4xx errors from compress instead of retrying (#4751) (e602e4c)
  • smart-crusher: keep container-shaped strings as strings in lossless tables (#4875) (303f521)
  • sql: stop granting anon read access to raw proxy telemetry (#4867) (b324efb)
  • tag-protector: make tag protection and restore linear in input size (#4841) (c8b3d67)
  • wrap: match Serena's unset dashboard key instead of appending a duplicate (#4315) (1da8930)

Performance Improvements

  • bm25: lower batch allocations -- 93.4% less peak memory (#4893) (cb8414c)

Dependencies

  • bump pyo3 from 0.29.3 to 0.30.1 (#4811) (5df3d94)
  • bump ruff from 0.17.3 to 0.17.4 in the pip-minor-patch group (#4804) (e89d2c2)
  • bump the npm-minor-patch group across 3 directories with 10 updates (#4813) (6182b5f)

Want more savings? Headroom Enterprise saves more for teams and enterprises. Talk to us at hello@headroomlabs.ai or headroomlabs.ai.

Don't miss a new headroom release

NewReleases is sending notifications on new releases.