chore: release main (#2792)

🤖 I have created a release *beep* *boop*
---


<details><summary>0.35.0</summary>

##
[0.35.0](https://github.com/headroomlabs-ai/headroom/compare/v0.34.0...v0.35.0)
(2026-08-12)


### Features

* **beacon:** allowlist the routing summary key
([#2818](https://github.com/headroomlabs-ai/headroom/issues/2818))
([7940c05](7940c05ebf))
* **beacon:** hourly R2 compaction, per-strategy savings, and a stack
that reports
([#2853](https://github.com/headroomlabs-ai/headroom/issues/2853))
([e0870ef](e0870ef931))
* **cli,pricing:** add CLI extension seam and prompt-cache TTL pricing
([#2802](https://github.com/headroomlabs-ai/headroom/issues/2802))
([6ec3e34](6ec3e3478a))


### Bug Fixes

* **anthropic:** strip first-party tool search on custom upstreams
([#2539](https://github.com/headroomlabs-ai/headroom/issues/2539))
([7f6950b](7f6950be34))
* **backends/anyllm:** convert Anthropic tools and tool_choice to OpenAI
shape
([0d6866b](0d6866b91a))
* **backends/anyllm:** stream tool_use blocks and map finish_reason on
the streaming path
([e4904e2](e4904e23a6))
* **backends/litellm:** None-guard core token counts in OpenAI usage
block ([#2324](https://github.com/headroomlabs-ai/headroom/issues/2324))
([12f9f58](12f9f58cb3))
* **beacon:** report all-layers savings, not context-compression only
([#2796](https://github.com/headroomlabs-ai/headroom/issues/2796))
([e9a24f3](e9a24f3ec1))
* **beacon:** split session failures by status code
([#2815](https://github.com/headroomlabs-ai/headroom/issues/2815))
([2954e37](2954e37048))
* **cache:** bound compression cache bookkeeping
([0ae948c](0ae948c151))
* **cache:** enforce Anthropic's 1h-before-5m cache_control ordering
before forwarding
([#2941](https://github.com/headroomlabs-ai/headroom/issues/2941))
([3752458](3752458022))
* **cache:** mirror client cache_control positions instead of
single-marker consolidation
([def3d76](def3d76e5a))
* **cache:** stabilize Anthropic block-growing lineages
([#2917](https://github.com/headroomlabs-ai/headroom/issues/2917))
([1a04c95](1a04c957f5))
* **ccr:** avoid injecting tool on chat streaming
([d0c1f5b](d0c1f5b8ad))
* **ccr:** preserve exact SQLite TTL boundary
([#2669](https://github.com/headroomlabs-ai/headroom/issues/2669))
([d0a86d4](d0a86d409f))
* **ccr:** report embedded hashes from compress endpoint
([#717](https://github.com/headroomlabs-ai/headroom/issues/717))
([685ebe4](685ebe457d))
* **ccr:** resolve &lt;&lt;ccr:...&gt;&gt; markers inline when no
retrieve-tool path exists
([#2512](https://github.com/headroomlabs-ai/headroom/issues/2512))
([ce8ce83](ce8ce8313f))
* **ccr:** tolerate null/malformed OpenAI data in response handling
([#2467](https://github.com/headroomlabs-ai/headroom/issues/2467))
([e583e08](e583e082d8))
* **ci:** publish latest from the root Docker manifest
([#2252](https://github.com/headroomlabs-ai/headroom/issues/2252))
([5568d73](5568d738af))
* **claude:** stop forcing tool search on Foundry
([#2477](https://github.com/headroomlabs-ai/headroom/issues/2477))
([7981396](798139608c))
* **cli/update:** let install ownership win over bare /.dockerenv so
venv installs self-update
([#2830](https://github.com/headroomlabs-ai/headroom/issues/2830))
([7092b53](7092b53c46))
* **codex:** route alpha search through the Codex backend
([#2538](https://github.com/headroomlabs-ai/headroom/issues/2538))
([a540eb2](a540eb2c61))
* **content-router:** protect custom-tag blocks before mixed-content
section split
([d7bc1e2](d7bc1e275f))
* **deps:** bump h2 to 4.4.1 for CVE-2026-71554
([#2839](https://github.com/headroomlabs-ai/headroom/issues/2839))
([564e0a8](564e0a8d0f))
* **deps:** enforce audited transitive dependency floors
([#2791](https://github.com/headroomlabs-ai/headroom/issues/2791))
([64e2039](64e203931b))
* **doctor:** flag `ollama launch claude` proxy bypass instead of
misdirecting
([#2566](https://github.com/headroomlabs-ai/headroom/issues/2566))
([7f24d69](7f24d695ee))
* emit SSE ping before message_start on Bedrock streaming path (issue
[#902](https://github.com/headroomlabs-ai/headroom/issues/902))
([#1080](https://github.com/headroomlabs-ai/headroom/issues/1080))
([4dab254](4dab254d52))
* **gemini:** resolve native CCR retrieval calls
([#2253](https://github.com/headroomlabs-ai/headroom/issues/2253))
([2483f57](2483f57002))
* **health:** label kompress as degraded/optional when not yet loaded
([#2865](https://github.com/headroomlabs-ai/headroom/issues/2865))
([8949371](89493714d2))
* **image:** decouple routing types from trained_router so importing the
compressor doesn't import torch
([#2513](https://github.com/headroomlabs-ai/headroom/issues/2513))
([#2537](https://github.com/headroomlabs-ai/headroom/issues/2537))
([d7cf981](d7cf981093))
* **install/windows:** register persistent-task from S4U hidden XML
([#2453](https://github.com/headroomlabs-ai/headroom/issues/2453))
([#2459](https://github.com/headroomlabs-ai/headroom/issues/2459))
([1edaeb8](1edaeb8b76))
* **install:** don't crash the PowerShell installer when $PROFILE is
unset ([#2469](https://github.com/headroomlabs-ai/headroom/issues/2469))
([fc5c4e2](fc5c4e239c))
* **install:** trust Docker bridge for dashboard metadata
([e044139](e044139001))
* **install:** use --userns=keep-id under Podman so bind-mount writes
don't fail
([#2846](https://github.com/headroomlabs-ai/headroom/issues/2846))
([3488f8d](3488f8d4b5))
* **learn/gemini:** stop double-counting session tokens
([#2230](https://github.com/headroomlabs-ai/headroom/issues/2230))
([29d8a5e](29d8a5e563))
* **learn/grok:** detect a Windows absolute project path
([#2283](https://github.com/headroomlabs-ai/headroom/issues/2283))
([e240df2](e240df2b69))
* **learn:** stop classifying a successful exit code 0 as an error
([#2289](https://github.com/headroomlabs-ai/headroom/issues/2289))
([a24fe7d](a24fe7dcbf))
* **litellm:** add async_post_call_success_hook to HeadroomCallback
([#1322](https://github.com/headroomlabs-ai/headroom/issues/1322))
([3107994](3107994aed))
* **litellm:** don't forward a caller key the target cannot accept
([#2883](https://github.com/headroomlabs-ai/headroom/issues/2883))
([2f2950a](2f2950a626))
* **memory:** bound the TrafficLearner pending-pattern accumulator
(memory leak)
([#2579](https://github.com/headroomlabs-ai/headroom/issues/2579))
([1f5feff](1f5fefffd3))
* **memory:** close DirectMem0 resources
([6596182](65961827cf))
* **memory:** close MCP backend on shutdown
([4bd8ecd](4bd8ecd1e3))
* **memory:** don't crash inline memory extraction on a non-object
&lt;memory&gt; block
([#2470](https://github.com/headroomlabs-ai/headroom/issues/2470))
([e00c6ff](e00c6ff81c))
* **memory:** keep vector metadata in sync
([#2295](https://github.com/headroomlabs-ai/headroom/issues/2295))
([c471800](c471800e8e))
* **memory:** make explicit-project and user store keys
collision-resistant
([#2231](https://github.com/headroomlabs-ai/headroom/issues/2231))
([f840d5f](f840d5f2fe))
* **memory:** skip &lt;system-reminder&gt; blocks when building the
retrieval query
([#2195](https://github.com/headroomlabs-ai/headroom/issues/2195))
([#2541](https://github.com/headroomlabs-ai/headroom/issues/2541))
([4e5a67a](4e5a67a342))
* **memory:** sync FTS5 and vector indexes on CLI
delete/edit/prune/purge
([fd4628d](fd4628d821))
* **oauth2:** make repository lint checks pass
([c85abf7](c85abf7a87))
* **observability:** aggregate tool savings in OTEL
([#2936](https://github.com/headroomlabs-ai/headroom/issues/2936))
([941c25d](941c25d31e))
* **onnx:** stop ONNX thread pools from spinning idle cores
([#2495](https://github.com/headroomlabs-ai/headroom/issues/2495))
([#2540](https://github.com/headroomlabs-ai/headroom/issues/2540))
([5c561bd](5c561bd913))
* **openai:** skip Responses tool-search deferral for clients that
cannot execute it
([#2696](https://github.com/headroomlabs-ai/headroom/issues/2696))
([54ea28d](54ea28d983))
* **opencode:** ship the transport hook-shim so wheel installs route
Node child traffic
([702dbc5](702dbc5902))
* **providers/anthropic:** don't crash token estimation on null
tool_calls
([#2472](https://github.com/headroomlabs-ai/headroom/issues/2472))
([08466f3](08466f3cae))
* **providers/openai:** bound tiktoken vocab loads with the guarded
loader
([#2554](https://github.com/headroomlabs-ai/headroom/issues/2554))
([0805e8e](0805e8e410))
* **proxy/anthropic:** inject headroom_retrieve whenever a CCR marker is
present, not only for new markers
([#2848](https://github.com/headroomlabs-ai/headroom/issues/2848))
([3808f60](3808f60ca6))
* **proxy/anthropic:** None-guard usage token counts on the direct
buffered path
([#2434](https://github.com/headroomlabs-ai/headroom/issues/2434))
([2b5ee7c](2b5ee7cde8))
* **proxy/anthropic:** run tool-search history repair after turn hooks
([c6f9948](c6f99482e1))
* **proxy/batch:** don't crash an OpenAI batch on a valid-JSON
non-object line
([#2316](https://github.com/headroomlabs-ai/headroom/issues/2316))
([1f2c681](1f2c681c0b))
* **proxy/bedrock:** report uncached input tokens from backend usage,
not the live-zone count
([#2318](https://github.com/headroomlabs-ai/headroom/issues/2318))
([c19e412](c19e412b33))
* **proxy/gemini:** keep streaming-parity baseline so eligible_pct can't
exceed 100
([#2824](https://github.com/headroomlabs-ai/headroom/issues/2824))
([b97c7c6](b97c7c6e99))
* **proxy/metrics:** cap client-supplied model label cardinality
([#2480](https://github.com/headroomlabs-ai/headroom/issues/2480))
([e24a7e6](e24a7e66b9))
* **proxy/metrics:** escape label values in the Prometheus export
([#2463](https://github.com/headroomlabs-ai/headroom/issues/2463))
([6a53861](6a53861063))
* **proxy/openai:** don't crash the Responses memory tool loops on null
arguments
([#2273](https://github.com/headroomlabs-ai/headroom/issues/2273))
([a30db2c](a30db2cae4))
* **proxy/openai:** feed Codex WS traffic into the traffic learner
([#2334](https://github.com/headroomlabs-ai/headroom/issues/2334))
([f669149](f669149769))
* **proxy/openai:** run response hooks on Responses, and bill their
re-drives
([#2872](https://github.com/headroomlabs-ai/headroom/issues/2872))
([675d13f](675d13f08d))
* **proxy:** allow settings routes for trusted gateway/dashboard clients
([#2491](https://github.com/headroomlabs-ai/headroom/issues/2491))
([a5b0a8f](a5b0a8f4cc))
* **proxy:** cache litellm model resolution to stop repeated Provider
List spam
([99f07e7](99f07e7bbd))
* **proxy:** cancel periodic TOIN task on shutdown
([739fdef](739fdef423))
* **proxy:** close the upstream stream when a streaming body is never
consumed
([0951663](0951663562))
* **proxy:** compress cache-mode cold starts and tag prefix-mismatch
passthrough
([#2365](https://github.com/headroomlabs-ai/headroom/issues/2365))
([aaeba0a](aaeba0a319))
* **proxy:** emit request log timestamps in UTC
([620028f](620028fa18))
* **proxy:** enable tool search by default and repair poisoned
transcripts
([#2807](https://github.com/headroomlabs-ai/headroom/issues/2807))
([0237cbf](0237cbffbb))
* **proxy:** gate mid-turn message coalescing to Claude Code clients
([#1643](https://github.com/headroomlabs-ai/headroom/issues/1643))
([a4bd2e6](a4bd2e62a5))
* **proxy:** give each Codex /v1/responses WS turn a unique request_id
([#2164](https://github.com/headroomlabs-ai/headroom/issues/2164))
([d02df10](d02df10758))
* **proxy:** graceful shutdown and reliable Ctrl+C exit
([#621](https://github.com/headroomlabs-ai/headroom/issues/621))
([17cdb18](17cdb185bc))
* **proxy:** guard telemetry and TOIN endpoints
([cde1513](cde1513c91))
* **proxy:** include tool_search_deferral savings in the savings ledger
([12149f7](12149f7446))
* **proxy:** pass through cross-region prefixed Bedrock model IDs
directly
([#2330](https://github.com/headroomlabs-ai/headroom/issues/2330))
([64cb46e](64cb46e24b))
* **proxy:** port session-sticky beta headers to the Rust proxy
([#2381](https://github.com/headroomlabs-ai/headroom/issues/2381))
([f6398a6](f6398a6476))
* **proxy:** preserve merged session and quarantine contracts
([#2943](https://github.com/headroomlabs-ai/headroom/issues/2943))
([039cd24](039cd2431a))
* **proxy:** preserve signed Anthropic thinking blocks on outbound
re-serialize
([#2254](https://github.com/headroomlabs-ai/headroom/issues/2254))
([dc163bc](dc163bcd1c))
* **proxy:** stop discarding compressed Codex WS later-frame payloads
([#2823](https://github.com/headroomlabs-ai/headroom/issues/2823))
([4ec416d](4ec416df88))
* **proxy:** time-cap the compression timeout-debt quarantine
([#2360](https://github.com/headroomlabs-ai/headroom/issues/2360))
([#2412](https://github.com/headroomlabs-ai/headroom/issues/2412))
([c5a08d2](c5a08d22e0))
* **proxy:** unwrap Hermes tool_call bridge in tool name map
([#2717](https://github.com/headroomlabs-ai/headroom/issues/2717))
([a97b824](a97b82413b))
* publish headroom-opencode in release workflow
([#2372](https://github.com/headroomlabs-ai/headroom/issues/2372))
([7859154](78591545ce))
* **settings:** accept documented HEADROOM_* env names as settings keys
([#2833](https://github.com/headroomlabs-ai/headroom/issues/2833))
([de9e052](de9e0523da))
* **subscription:** dedup transcript usage by message id
([#2340](https://github.com/headroomlabs-ai/headroom/issues/2340) token
inflation)
([#2408](https://github.com/headroomlabs-ai/headroom/issues/2408))
([74275b7](74275b7c3e))
* **toin:** bound private query and pattern retention
([8cd1380](8cd138039e))
* **tokenizer:** coerce non-string tool_call fields before counting
([#2801](https://github.com/headroomlabs-ai/headroom/issues/2801))
([b6f9877](b6f9877c78))
* **tokenizer:** price CJK in the Rust fixed-ratio estimator (Python
parity)
([#2260](https://github.com/headroomlabs-ai/headroom/issues/2260))
([6840153](6840153473))
* **transforms/adaptive-sizer:** honor max_k on small-input fast path
([#2319](https://github.com/headroomlabs-ai/headroom/issues/2319))
([8a90523](8a90523209))
* **transforms/smart_crusher:** don't crash on a tool call with a null
function
([#2232](https://github.com/headroomlabs-ai/headroom/issues/2232))
([3bb02f8](3bb02f8f75))
* Vertex model pricing shows $0.00 for versioned model names and
vertex:anthropic provider
([#2517](https://github.com/headroomlabs-ai/headroom/issues/2517))
([eb5b5e4](eb5b5e4198))
* **wrap/claude:** keep --1m effective when an explicit --model is
passed through
([c093bf1](c093bf11eb))
* **wrap/opencode:** verify the opencode binary before mutating config
([ae38486](ae384862a4))
* **wrap/serena:** install Serena from the serena-agent PyPI wheel, not
the git source
([d7b25ae](d7b25ae3bb))
* **wrap:** honor Copilot OAuth wire-api override and model default
([#2387](https://github.com/headroomlabs-ai/headroom/issues/2387))
([1db6d88](1db6d88ab4))
* **wrap:** serialize shared proxy startup
([#2946](https://github.com/headroomlabs-ai/headroom/issues/2946))
([e540d64](e540d64feb))
* **wrap:** stop the launch cwd from shadowing the installed package in
the proxy subprocess
([#2843](https://github.com/headroomlabs-ai/headroom/issues/2843))
([c49be26](c49be269a1))


### Performance Improvements

* cut hot-path latency 27% (token-count memo, startup preloads, JSON
scan memo)
([#2838](https://github.com/headroomlabs-ai/headroom/issues/2838))
([53af90d](53af90d68c))
* **proxy:** bound upstream calls and hot-path costs
([#2852](https://github.com/headroomlabs-ai/headroom/issues/2852))
([f624d3a](f624d3a00a))
* **subscription:** skip transcripts older than the window in
compute_window_tokens
([#2861](https://github.com/headroomlabs-ai/headroom/issues/2861))
([91d6bf3](91d6bf33cd))


### Dependencies

* bump brace-expansion from 5.0.7 to 5.0.9 in /docs
([#2751](https://github.com/headroomlabs-ai/headroom/issues/2751))
([56ee57b](56ee57be98))
* bump bytesize from 1.3.3 to 2.4.2
([#2286](https://github.com/headroomlabs-ai/headroom/issues/2286))
([6448545](6448545a7f))
* bump hf-hub from 0.4.3 to 0.5.0
([#2285](https://github.com/headroomlabs-ai/headroom/issues/2285))
([4925bf6](4925bf6a82))
* bump next from 16.2.10 to 16.3.0 in /docs
([#2750](https://github.com/headroomlabs-ai/headroom/issues/2750))
([0fd0b99](0fd0b996a4))
* bump postcss from 8.5.19 to 8.5.25 in /plugins/openclaw
([#2749](https://github.com/headroomlabs-ai/headroom/issues/2749))
([cd60ee9](cd60ee9ae8))
* bump postcss from 8.5.19 to 8.5.25 in /plugins/opencode
([#2748](https://github.com/headroomlabs-ai/headroom/issues/2748))
([ff4e016](ff4e0167bb))
* bump postcss from 8.5.19 to 8.5.25 in /sdk/typescript
([#2747](https://github.com/headroomlabs-ai/headroom/issues/2747))
([267c2bd](267c2bdcb5))
* bump postcss from 8.5.19 to 8.5.26 in /docs
([#2881](https://github.com/headroomlabs-ai/headroom/issues/2881))
([e6e5826](e6e5826423))
* bump ruff from 0.15.17 to 0.15.22 in the pip-minor-patch group
([#2501](https://github.com/headroomlabs-ai/headroom/issues/2501))
([ecf130d](ecf130d3ac))
* bump rusqlite from 0.32.1 to 0.40.1
([#2287](https://github.com/headroomlabs-ai/headroom/issues/2287))
([522faa1](522faa1a59))
* bump the cargo-minor-patch group across 1 directory with 22 updates
([#2916](https://github.com/headroomlabs-ai/headroom/issues/2916))
([148d860](148d8605e2))
</details>

---
This PR was generated with [Release
Please](https://github.com/googleapis/release-please). See
[documentation](https://github.com/googleapis/release-please#release-please).

---------

Co-authored-by: JD Davis <mxjerrett@gmail.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

.claude-plugin/marketplace.json

@@ -5,14 +5,14 @@
   },
   "metadata": {
     "description": "Headroom marketplace for Claude Code and GitHub Copilot CLI plugins.",
-    "version": "0.34.0"
+    "version": "0.35.0"
   },
   "plugins": [
     {
       "name": "headroom",
       "source": "./plugins/headroom-agent-hooks",
       "description": "Headroom startup hooks for Claude Code and GitHub Copilot CLI.",
-      "version": "0.34.0",
+      "version": "0.35.0",
       "author": {
         "name": "Headroom Contributors",
         "url": "https://github.com/chopratejas/headroom"

.github/plugin/marketplace.json

@@ -5,14 +5,14 @@
   },
   "metadata": {
     "description": "Headroom marketplace for Claude Code and GitHub Copilot CLI plugins.",
-    "version": "0.34.0"
+    "version": "0.35.0"
   },
   "plugins": [
     {
       "name": "headroom",
       "source": "./plugins/headroom-agent-hooks",
       "description": "Headroom startup hooks for Claude Code and GitHub Copilot CLI.",
-      "version": "0.34.0",
+      "version": "0.35.0",
       "author": {
         "name": "Headroom Contributors",
         "url": "https://github.com/chopratejas/headroom"

.release-please-manifest.json

@@ -1,3 +1,3 @@
 {
-  ".": "0.34.0"
+  ".": "0.35.0"
 }

.releasemetadata

@@ -1,9 +1,10 @@
 {
-  "version": "0.34.0",
+  "version": "0.35.0",
   "packages": {
-    "pypi": "0.34.0",
+    "pypi": "0.35.0",
-    "npm-sdk": "0.34.0",
+    "npm-sdk": "0.35.0",
-    "npm-openclaw": "0.34.0",
+    "npm-openclaw": "0.35.0",
-    "agent-hooks-plugin": "0.34.0"
+    "npm-opencode": "0.35.0",
+    "agent-hooks-plugin": "0.35.0"
   }
 }

CHANGELOG.md

@@ -284,6 +284,137 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - **code:** fix two `CodeAwareCompressor` AST-reassembly bugs: an exported JS/TS function or class (`export function foo() {`) produced a duplicated `export export` keyword and invalid syntax, because line-based node slicing (used to preserve indentation) pulled in the preceding `export` sibling's text on top of the `export_statement` handler's own prefix reconstruction. Separately, in every supported language, a doc comment immediately above a top-level function, class, or type was detached from its declaration during extraction and re-emitted in a cluster at the end of the compressed output instead of staying attached to what it documents.
 - * **proxy:** Buffered upstream responses containing a `server_tool_use` (or any other unrecognized Anthropic content block) no longer turn a fully-generated response into an HTTP 502. `StreamingMixin._response_to_sse` raised `ValueError` on unknown block types after the entire upstream generation had already been buffered, so a slow-but-successful response failed and the client retried the whole multi-minute request. Unknown blocks are now emitted verbatim in `content_block_start` (following the existing redacted_thinking` pattern), so `server_tool_use`, `server_tool_result`, `mcp_tool_use`, and future block types round-trip ([#1806](https://github.com/headroomlabs-ai/headroom/issues/1806)).
 
+## [0.35.0](https://github.com/headroomlabs-ai/headroom/compare/v0.34.0...v0.35.0) (2026-08-12)
+
+
+### Features
+
+* **beacon:** allowlist the routing summary key ([#2818](https://github.com/headroomlabs-ai/headroom/issues/2818)) ([7940c05](https://github.com/headroomlabs-ai/headroom/commit/7940c05ebf4486c6b9d00984067ae33cedf4dddb))
+* **beacon:** hourly R2 compaction, per-strategy savings, and a stack that reports ([#2853](https://github.com/headroomlabs-ai/headroom/issues/2853)) ([e0870ef](https://github.com/headroomlabs-ai/headroom/commit/e0870ef931e5ea6cc6cb52551f5d80cd9e3dc715))
+* **cli,pricing:** add CLI extension seam and prompt-cache TTL pricing ([#2802](https://github.com/headroomlabs-ai/headroom/issues/2802)) ([6ec3e34](https://github.com/headroomlabs-ai/headroom/commit/6ec3e3478abf058fe1460f91342bcdadf54a1ba8))
+
+
+### Bug Fixes
+
+* **anthropic:** strip first-party tool search on custom upstreams ([#2539](https://github.com/headroomlabs-ai/headroom/issues/2539)) ([7f6950b](https://github.com/headroomlabs-ai/headroom/commit/7f6950be34e29304deae0fa5138b852491b092fe))
+* **backends/anyllm:** convert Anthropic tools and tool_choice to OpenAI shape ([0d6866b](https://github.com/headroomlabs-ai/headroom/commit/0d6866b91a3777475abd58cd8b63a10cd0621e7f))
+* **backends/anyllm:** stream tool_use blocks and map finish_reason on the streaming path ([e4904e2](https://github.com/headroomlabs-ai/headroom/commit/e4904e23a6ba6f5cff2488332946481172446922))
+* **backends/litellm:** None-guard core token counts in OpenAI usage block ([#2324](https://github.com/headroomlabs-ai/headroom/issues/2324)) ([12f9f58](https://github.com/headroomlabs-ai/headroom/commit/12f9f58cb3dcfc67af1238424d404d8dd9bad1dd))
+* **beacon:** report all-layers savings, not context-compression only ([#2796](https://github.com/headroomlabs-ai/headroom/issues/2796)) ([e9a24f3](https://github.com/headroomlabs-ai/headroom/commit/e9a24f3ec1ffd278b0b3ca547a90942c40c99ec8))
+* **beacon:** split session failures by status code ([#2815](https://github.com/headroomlabs-ai/headroom/issues/2815)) ([2954e37](https://github.com/headroomlabs-ai/headroom/commit/2954e37048f8dcffe16e1c37b8f71afb0094a0a2))
+* **cache:** bound compression cache bookkeeping ([0ae948c](https://github.com/headroomlabs-ai/headroom/commit/0ae948c1510735df39317bf0861f8a8750cdbf9d))
+* **cache:** enforce Anthropic's 1h-before-5m cache_control ordering before forwarding ([#2941](https://github.com/headroomlabs-ai/headroom/issues/2941)) ([3752458](https://github.com/headroomlabs-ai/headroom/commit/3752458022f736c779f7b5a6c2d6d2ef0bc89f72))
+* **cache:** mirror client cache_control positions instead of single-marker consolidation ([def3d76](https://github.com/headroomlabs-ai/headroom/commit/def3d76e5ab4665e609b51bfba54dd6d25116925))
+* **cache:** stabilize Anthropic block-growing lineages ([#2917](https://github.com/headroomlabs-ai/headroom/issues/2917)) ([1a04c95](https://github.com/headroomlabs-ai/headroom/commit/1a04c957f53ef25ab1209166f425a7876913c4d3))
+* **ccr:** avoid injecting tool on chat streaming ([d0c1f5b](https://github.com/headroomlabs-ai/headroom/commit/d0c1f5b8ad68c7a44ed3aaa0fe40e3a656950123))
+* **ccr:** preserve exact SQLite TTL boundary ([#2669](https://github.com/headroomlabs-ai/headroom/issues/2669)) ([d0a86d4](https://github.com/headroomlabs-ai/headroom/commit/d0a86d409fab377f9c642d1f3680b6ece7f97b8a))
+* **ccr:** report embedded hashes from compress endpoint ([#717](https://github.com/headroomlabs-ai/headroom/issues/717)) ([685ebe4](https://github.com/headroomlabs-ai/headroom/commit/685ebe457d727922ba4057515556a2d2aac0f616))
+* **ccr:** resolve <<ccr:...>> markers inline when no retrieve-tool path exists ([#2512](https://github.com/headroomlabs-ai/headroom/issues/2512)) ([ce8ce83](https://github.com/headroomlabs-ai/headroom/commit/ce8ce8313f8cebf060392a62f9adaab18c0df386))
+* **ccr:** tolerate null/malformed OpenAI data in response handling ([#2467](https://github.com/headroomlabs-ai/headroom/issues/2467)) ([e583e08](https://github.com/headroomlabs-ai/headroom/commit/e583e082d8dee942229ac6211c742f9c9448a905))
+* **ci:** publish latest from the root Docker manifest ([#2252](https://github.com/headroomlabs-ai/headroom/issues/2252)) ([5568d73](https://github.com/headroomlabs-ai/headroom/commit/5568d738afb5e080d8df56e64500026996cbf025))
+* **claude:** stop forcing tool search on Foundry ([#2477](https://github.com/headroomlabs-ai/headroom/issues/2477)) ([7981396](https://github.com/headroomlabs-ai/headroom/commit/798139608c0fb5118eb3a7a183b8b2abe92341f1))
+* **cli/update:** let install ownership win over bare /.dockerenv so venv installs self-update ([#2830](https://github.com/headroomlabs-ai/headroom/issues/2830)) ([7092b53](https://github.com/headroomlabs-ai/headroom/commit/7092b53c466bf5dbda8a1cda88403d1a4b16deb1))
+* **codex:** route alpha search through the Codex backend ([#2538](https://github.com/headroomlabs-ai/headroom/issues/2538)) ([a540eb2](https://github.com/headroomlabs-ai/headroom/commit/a540eb2c61b1a47e5ab8b07ea4a80fee780b6514))
+* **content-router:** protect custom-tag blocks before mixed-content section split ([d7bc1e2](https://github.com/headroomlabs-ai/headroom/commit/d7bc1e275f411788abffa2d007db14aa17fd31c5))
+* **deps:** bump h2 to 4.4.1 for CVE-2026-71554 ([#2839](https://github.com/headroomlabs-ai/headroom/issues/2839)) ([564e0a8](https://github.com/headroomlabs-ai/headroom/commit/564e0a8d0fe440dff21a6c405c88e05698b3059f))
+* **deps:** enforce audited transitive dependency floors ([#2791](https://github.com/headroomlabs-ai/headroom/issues/2791)) ([64e2039](https://github.com/headroomlabs-ai/headroom/commit/64e203931b9810e5a010f063d26d154419016f86))
+* **doctor:** flag `ollama launch claude` proxy bypass instead of misdirecting ([#2566](https://github.com/headroomlabs-ai/headroom/issues/2566)) ([7f24d69](https://github.com/headroomlabs-ai/headroom/commit/7f24d695eea00b9bb3265fbaa6629acf0c2ff181))
+* emit SSE ping before message_start on Bedrock streaming path (issue [#902](https://github.com/headroomlabs-ai/headroom/issues/902)) ([#1080](https://github.com/headroomlabs-ai/headroom/issues/1080)) ([4dab254](https://github.com/headroomlabs-ai/headroom/commit/4dab254d52914c39ffe13071848604e1771b1bd1))
+* **gemini:** resolve native CCR retrieval calls ([#2253](https://github.com/headroomlabs-ai/headroom/issues/2253)) ([2483f57](https://github.com/headroomlabs-ai/headroom/commit/2483f570025763cd9183a93749ea8cf38f1aeb85))
+* **health:** label kompress as degraded/optional when not yet loaded ([#2865](https://github.com/headroomlabs-ai/headroom/issues/2865)) ([8949371](https://github.com/headroomlabs-ai/headroom/commit/89493714d2cffdc1f81a8f417ea09891453d7009))
+* **image:** decouple routing types from trained_router so importing the compressor doesn't import torch ([#2513](https://github.com/headroomlabs-ai/headroom/issues/2513)) ([#2537](https://github.com/headroomlabs-ai/headroom/issues/2537)) ([d7cf981](https://github.com/headroomlabs-ai/headroom/commit/d7cf981093cf505192a3736dadd0254a120830a1))
+* **install/windows:** register persistent-task from S4U hidden XML ([#2453](https://github.com/headroomlabs-ai/headroom/issues/2453)) ([#2459](https://github.com/headroomlabs-ai/headroom/issues/2459)) ([1edaeb8](https://github.com/headroomlabs-ai/headroom/commit/1edaeb8b76f6b872a6c810d404c944caf1a594b2))
+* **install:** don't crash the PowerShell installer when $PROFILE is unset ([#2469](https://github.com/headroomlabs-ai/headroom/issues/2469)) ([fc5c4e2](https://github.com/headroomlabs-ai/headroom/commit/fc5c4e239ce32f2b90a6777772a01bdf49c66cb6))
+* **install:** trust Docker bridge for dashboard metadata ([e044139](https://github.com/headroomlabs-ai/headroom/commit/e044139001680fd5198147bf373df6f00db32cc7))
+* **install:** use --userns=keep-id under Podman so bind-mount writes don't fail ([#2846](https://github.com/headroomlabs-ai/headroom/issues/2846)) ([3488f8d](https://github.com/headroomlabs-ai/headroom/commit/3488f8d4b5fae4eab157e0c4031ccf712bcbcc0d))
+* **learn/gemini:** stop double-counting session tokens ([#2230](https://github.com/headroomlabs-ai/headroom/issues/2230)) ([29d8a5e](https://github.com/headroomlabs-ai/headroom/commit/29d8a5e563cf16dbd3a53a1571f4f352e61e1b33))
+* **learn/grok:** detect a Windows absolute project path ([#2283](https://github.com/headroomlabs-ai/headroom/issues/2283)) ([e240df2](https://github.com/headroomlabs-ai/headroom/commit/e240df2b698e601324b85956bd93cb304f6030ab))
+* **learn:** stop classifying a successful exit code 0 as an error ([#2289](https://github.com/headroomlabs-ai/headroom/issues/2289)) ([a24fe7d](https://github.com/headroomlabs-ai/headroom/commit/a24fe7dcbfe5ab30d0cef631c936e2245c12d123))
+* **litellm:** add async_post_call_success_hook to HeadroomCallback ([#1322](https://github.com/headroomlabs-ai/headroom/issues/1322)) ([3107994](https://github.com/headroomlabs-ai/headroom/commit/3107994aed5fd42e713d3c26f3f08121a62b980e))
+* **litellm:** don't forward a caller key the target cannot accept ([#2883](https://github.com/headroomlabs-ai/headroom/issues/2883)) ([2f2950a](https://github.com/headroomlabs-ai/headroom/commit/2f2950a626cebf851aac29255e7188fbb1639f5a))
+* **memory:** bound the TrafficLearner pending-pattern accumulator (memory leak) ([#2579](https://github.com/headroomlabs-ai/headroom/issues/2579)) ([1f5feff](https://github.com/headroomlabs-ai/headroom/commit/1f5fefffd3e82c73bddd928cfd53334031e807bc))
+* **memory:** close DirectMem0 resources ([6596182](https://github.com/headroomlabs-ai/headroom/commit/65961827cf5e90d7b4e7026feb89aac000a73ea3))
+* **memory:** close MCP backend on shutdown ([4bd8ecd](https://github.com/headroomlabs-ai/headroom/commit/4bd8ecd1e31475365801791d35630f66f7393553))
+* **memory:** don't crash inline memory extraction on a non-object <memory> block ([#2470](https://github.com/headroomlabs-ai/headroom/issues/2470)) ([e00c6ff](https://github.com/headroomlabs-ai/headroom/commit/e00c6ff81ce2003e04042b8f2d1bd6aa3c6e885c))
+* **memory:** keep vector metadata in sync ([#2295](https://github.com/headroomlabs-ai/headroom/issues/2295)) ([c471800](https://github.com/headroomlabs-ai/headroom/commit/c471800e8ee22986c308464b02a85da5575f34cc))
+* **memory:** make explicit-project and user store keys collision-resistant ([#2231](https://github.com/headroomlabs-ai/headroom/issues/2231)) ([f840d5f](https://github.com/headroomlabs-ai/headroom/commit/f840d5f2fe938432e542c3f71f2218eeecd06b05))
+* **memory:** skip <system-reminder> blocks when building the retrieval query ([#2195](https://github.com/headroomlabs-ai/headroom/issues/2195)) ([#2541](https://github.com/headroomlabs-ai/headroom/issues/2541)) ([4e5a67a](https://github.com/headroomlabs-ai/headroom/commit/4e5a67a342be4be659b62c7863a9e72422605788))
+* **memory:** sync FTS5 and vector indexes on CLI delete/edit/prune/purge ([fd4628d](https://github.com/headroomlabs-ai/headroom/commit/fd4628d82156c65d4fa22df9513315790a6cd2fb))
+* **oauth2:** make repository lint checks pass ([c85abf7](https://github.com/headroomlabs-ai/headroom/commit/c85abf7a87920012e01f0a677f6fbd98c4b08de0))
+* **observability:** aggregate tool savings in OTEL ([#2936](https://github.com/headroomlabs-ai/headroom/issues/2936)) ([941c25d](https://github.com/headroomlabs-ai/headroom/commit/941c25d31e6c6e0b436c307cbe212771ff76b45f))
+* **onnx:** stop ONNX thread pools from spinning idle cores ([#2495](https://github.com/headroomlabs-ai/headroom/issues/2495)) ([#2540](https://github.com/headroomlabs-ai/headroom/issues/2540)) ([5c561bd](https://github.com/headroomlabs-ai/headroom/commit/5c561bd913ea60fad2c3c53f4b65e679e7d248d0))
+* **openai:** skip Responses tool-search deferral for clients that cannot execute it ([#2696](https://github.com/headroomlabs-ai/headroom/issues/2696)) ([54ea28d](https://github.com/headroomlabs-ai/headroom/commit/54ea28d9839a0dcfa4dd0cf4210a4421f03beeff))
+* **opencode:** ship the transport hook-shim so wheel installs route Node child traffic ([702dbc5](https://github.com/headroomlabs-ai/headroom/commit/702dbc5902ff184a7c20178958a811beb9c78fa3))
+* **providers/anthropic:** don't crash token estimation on null tool_calls ([#2472](https://github.com/headroomlabs-ai/headroom/issues/2472)) ([08466f3](https://github.com/headroomlabs-ai/headroom/commit/08466f3cae4dbb2647dc6f249fe42c4e840600c5))
+* **providers/openai:** bound tiktoken vocab loads with the guarded loader ([#2554](https://github.com/headroomlabs-ai/headroom/issues/2554)) ([0805e8e](https://github.com/headroomlabs-ai/headroom/commit/0805e8e410543d75c7ddd3b83dde5eda3bc13144))
+* **proxy/anthropic:** inject headroom_retrieve whenever a CCR marker is present, not only for new markers ([#2848](https://github.com/headroomlabs-ai/headroom/issues/2848)) ([3808f60](https://github.com/headroomlabs-ai/headroom/commit/3808f60ca61e84faf3ea8f8e003a6e6c8e9af4da))
+* **proxy/anthropic:** None-guard usage token counts on the direct buffered path ([#2434](https://github.com/headroomlabs-ai/headroom/issues/2434)) ([2b5ee7c](https://github.com/headroomlabs-ai/headroom/commit/2b5ee7cde809ca37f6998d9679b1eb2133ab50ca))
+* **proxy/anthropic:** run tool-search history repair after turn hooks ([c6f9948](https://github.com/headroomlabs-ai/headroom/commit/c6f99482e1bea024db6014a70c8e6da419543957))
+* **proxy/batch:** don't crash an OpenAI batch on a valid-JSON non-object line ([#2316](https://github.com/headroomlabs-ai/headroom/issues/2316)) ([1f2c681](https://github.com/headroomlabs-ai/headroom/commit/1f2c681c0b48150a569277d3ebd5e95709dc7c39))
+* **proxy/bedrock:** report uncached input tokens from backend usage, not the live-zone count ([#2318](https://github.com/headroomlabs-ai/headroom/issues/2318)) ([c19e412](https://github.com/headroomlabs-ai/headroom/commit/c19e412b3356d80dece001887d4ff48b6fd5150b))
+* **proxy/gemini:** keep streaming-parity baseline so eligible_pct can't exceed 100 ([#2824](https://github.com/headroomlabs-ai/headroom/issues/2824)) ([b97c7c6](https://github.com/headroomlabs-ai/headroom/commit/b97c7c6e99eac84df49c7a7e5f21dedb298716fe))
+* **proxy/metrics:** cap client-supplied model label cardinality ([#2480](https://github.com/headroomlabs-ai/headroom/issues/2480)) ([e24a7e6](https://github.com/headroomlabs-ai/headroom/commit/e24a7e66b95fa908c4ea6fd079809ece7692e6b2))
+* **proxy/metrics:** escape label values in the Prometheus export ([#2463](https://github.com/headroomlabs-ai/headroom/issues/2463)) ([6a53861](https://github.com/headroomlabs-ai/headroom/commit/6a53861063c3839e698bbec7194517bdfd851c38))
+* **proxy/openai:** don't crash the Responses memory tool loops on null arguments ([#2273](https://github.com/headroomlabs-ai/headroom/issues/2273)) ([a30db2c](https://github.com/headroomlabs-ai/headroom/commit/a30db2cae49b4ef03ebbd404ec1fc6c4f5f2404d))
+* **proxy/openai:** feed Codex WS traffic into the traffic learner ([#2334](https://github.com/headroomlabs-ai/headroom/issues/2334)) ([f669149](https://github.com/headroomlabs-ai/headroom/commit/f6691497692869b7067438597421ff12aace6bf4))
+* **proxy/openai:** run response hooks on Responses, and bill their re-drives ([#2872](https://github.com/headroomlabs-ai/headroom/issues/2872)) ([675d13f](https://github.com/headroomlabs-ai/headroom/commit/675d13f08d42455c8fa17bda878c1a11b905cee4))
+* **proxy:** allow settings routes for trusted gateway/dashboard clients ([#2491](https://github.com/headroomlabs-ai/headroom/issues/2491)) ([a5b0a8f](https://github.com/headroomlabs-ai/headroom/commit/a5b0a8f4cc54d68afcf371a422b3a4a9635b7e7f))
+* **proxy:** cache litellm model resolution to stop repeated Provider List spam ([99f07e7](https://github.com/headroomlabs-ai/headroom/commit/99f07e7bbdded9dadc70e35ee6ab025279d1aa22))
+* **proxy:** cancel periodic TOIN task on shutdown ([739fdef](https://github.com/headroomlabs-ai/headroom/commit/739fdef423fa8cbc82537481c875d4570b0ecad4))
+* **proxy:** close the upstream stream when a streaming body is never consumed ([0951663](https://github.com/headroomlabs-ai/headroom/commit/09516635621caccf7e3db4f537eb49ea49b8a453))
+* **proxy:** compress cache-mode cold starts and tag prefix-mismatch passthrough ([#2365](https://github.com/headroomlabs-ai/headroom/issues/2365)) ([aaeba0a](https://github.com/headroomlabs-ai/headroom/commit/aaeba0a319f12b98cad3bfcf1cf991b694b946bf))
+* **proxy:** emit request log timestamps in UTC ([620028f](https://github.com/headroomlabs-ai/headroom/commit/620028fa18843622d3e454bd40fb91a93e607dbf))
+* **proxy:** enable tool search by default and repair poisoned transcripts ([#2807](https://github.com/headroomlabs-ai/headroom/issues/2807)) ([0237cbf](https://github.com/headroomlabs-ai/headroom/commit/0237cbffbbc456ad8a7398005602d76881862d99))
+* **proxy:** gate mid-turn message coalescing to Claude Code clients ([#1643](https://github.com/headroomlabs-ai/headroom/issues/1643)) ([a4bd2e6](https://github.com/headroomlabs-ai/headroom/commit/a4bd2e62a5bb73f15b3b12e979c69e2b555bee10))
+* **proxy:** give each Codex /v1/responses WS turn a unique request_id ([#2164](https://github.com/headroomlabs-ai/headroom/issues/2164)) ([d02df10](https://github.com/headroomlabs-ai/headroom/commit/d02df1075894b414d60626aca2bbcadd7a3577a0))
+* **proxy:** graceful shutdown and reliable Ctrl+C exit ([#621](https://github.com/headroomlabs-ai/headroom/issues/621)) ([17cdb18](https://github.com/headroomlabs-ai/headroom/commit/17cdb185bc79d8cfec104e781a7e555af3ef11e1))
+* **proxy:** guard telemetry and TOIN endpoints ([cde1513](https://github.com/headroomlabs-ai/headroom/commit/cde1513c91b6c6c240869bc5660f4b8966197bbc))
+* **proxy:** include tool_search_deferral savings in the savings ledger ([12149f7](https://github.com/headroomlabs-ai/headroom/commit/12149f74466c08b69be8d5fe751425be63c2fda4))
+* **proxy:** pass through cross-region prefixed Bedrock model IDs directly ([#2330](https://github.com/headroomlabs-ai/headroom/issues/2330)) ([64cb46e](https://github.com/headroomlabs-ai/headroom/commit/64cb46e24bf7b223ea71b14b6f5e86e78fa7ac45))
+* **proxy:** port session-sticky beta headers to the Rust proxy ([#2381](https://github.com/headroomlabs-ai/headroom/issues/2381)) ([f6398a6](https://github.com/headroomlabs-ai/headroom/commit/f6398a64768a095b722a5fb0b2445c7953dee1c6))
+* **proxy:** preserve merged session and quarantine contracts ([#2943](https://github.com/headroomlabs-ai/headroom/issues/2943)) ([039cd24](https://github.com/headroomlabs-ai/headroom/commit/039cd2431aaec7d59fefaf7e97aeda1fd7ab3afa))
+* **proxy:** preserve signed Anthropic thinking blocks on outbound re-serialize ([#2254](https://github.com/headroomlabs-ai/headroom/issues/2254)) ([dc163bc](https://github.com/headroomlabs-ai/headroom/commit/dc163bcd1cba4cd8898f23286eb1365fcf6e0356))
+* **proxy:** stop discarding compressed Codex WS later-frame payloads ([#2823](https://github.com/headroomlabs-ai/headroom/issues/2823)) ([4ec416d](https://github.com/headroomlabs-ai/headroom/commit/4ec416df8899036544e679f561f1cf921f3da0dd))
+* **proxy:** time-cap the compression timeout-debt quarantine ([#2360](https://github.com/headroomlabs-ai/headroom/issues/2360)) ([#2412](https://github.com/headroomlabs-ai/headroom/issues/2412)) ([c5a08d2](https://github.com/headroomlabs-ai/headroom/commit/c5a08d22e05a7dd2b929f3cca76ee3fb42f122db))
+* **proxy:** unwrap Hermes tool_call bridge in tool name map ([#2717](https://github.com/headroomlabs-ai/headroom/issues/2717)) ([a97b824](https://github.com/headroomlabs-ai/headroom/commit/a97b82413bdc86655c064417ed4628ff4d9d7c9d))
+* publish headroom-opencode in release workflow ([#2372](https://github.com/headroomlabs-ai/headroom/issues/2372)) ([7859154](https://github.com/headroomlabs-ai/headroom/commit/78591545ceb8303fdf9b93cd5ff02b626df97d2b))
+* **settings:** accept documented HEADROOM_* env names as settings keys ([#2833](https://github.com/headroomlabs-ai/headroom/issues/2833)) ([de9e052](https://github.com/headroomlabs-ai/headroom/commit/de9e0523dad47b700062464adecd60f82547f332))
+* **subscription:** dedup transcript usage by message id ([#2340](https://github.com/headroomlabs-ai/headroom/issues/2340) token inflation) ([#2408](https://github.com/headroomlabs-ai/headroom/issues/2408)) ([74275b7](https://github.com/headroomlabs-ai/headroom/commit/74275b7c3e2b39be5198f9efa35057a5e026e665))
+* **toin:** bound private query and pattern retention ([8cd1380](https://github.com/headroomlabs-ai/headroom/commit/8cd138039edbfc295080ec474325d527fb3aedf3))
+* **tokenizer:** coerce non-string tool_call fields before counting ([#2801](https://github.com/headroomlabs-ai/headroom/issues/2801)) ([b6f9877](https://github.com/headroomlabs-ai/headroom/commit/b6f9877c78b3fa3b1d705426bd27d74be77f4fa0))
+* **tokenizer:** price CJK in the Rust fixed-ratio estimator (Python parity) ([#2260](https://github.com/headroomlabs-ai/headroom/issues/2260)) ([6840153](https://github.com/headroomlabs-ai/headroom/commit/6840153473caa0d61e982215e16a8cf54b0b6cc7))
+* **transforms/adaptive-sizer:** honor max_k on small-input fast path ([#2319](https://github.com/headroomlabs-ai/headroom/issues/2319)) ([8a90523](https://github.com/headroomlabs-ai/headroom/commit/8a905232091d993fac9e19a59bc449f201d4cdf3))
+* **transforms/smart_crusher:** don't crash on a tool call with a null function ([#2232](https://github.com/headroomlabs-ai/headroom/issues/2232)) ([3bb02f8](https://github.com/headroomlabs-ai/headroom/commit/3bb02f8f75f12cf8258a5b1c2a7fbdc190f9d074))
+* Vertex model pricing shows $0.00 for versioned model names and vertex:anthropic provider ([#2517](https://github.com/headroomlabs-ai/headroom/issues/2517)) ([eb5b5e4](https://github.com/headroomlabs-ai/headroom/commit/eb5b5e41988f5c27d29ae8ae3e5fe74e56493b8c))
+* **wrap/claude:** keep --1m effective when an explicit --model is passed through ([c093bf1](https://github.com/headroomlabs-ai/headroom/commit/c093bf11eb5f356f71367ebb7b56ae3c2b434a12))
+* **wrap/opencode:** verify the opencode binary before mutating config ([ae38486](https://github.com/headroomlabs-ai/headroom/commit/ae384862a4950cec057103e9daf75e74107640df))
+* **wrap/serena:** install Serena from the serena-agent PyPI wheel, not the git source ([d7b25ae](https://github.com/headroomlabs-ai/headroom/commit/d7b25ae3bb3364cde4931509ecb65e32085e5b09))
+* **wrap:** honor Copilot OAuth wire-api override and model default ([#2387](https://github.com/headroomlabs-ai/headroom/issues/2387)) ([1db6d88](https://github.com/headroomlabs-ai/headroom/commit/1db6d88ab4ea25654b8277358902b7df700db6b4))
+* **wrap:** serialize shared proxy startup ([#2946](https://github.com/headroomlabs-ai/headroom/issues/2946)) ([e540d64](https://github.com/headroomlabs-ai/headroom/commit/e540d64febf27f2e7997d3a1a1d89478cc1ef658))
+* **wrap:** stop the launch cwd from shadowing the installed package in the proxy subprocess ([#2843](https://github.com/headroomlabs-ai/headroom/issues/2843)) ([c49be26](https://github.com/headroomlabs-ai/headroom/commit/c49be269a18446779cd8a048caaa7f0ba3a3b48b))
+
+
+### Performance Improvements
+
+* cut hot-path latency 27% (token-count memo, startup preloads, JSON scan memo) ([#2838](https://github.com/headroomlabs-ai/headroom/issues/2838)) ([53af90d](https://github.com/headroomlabs-ai/headroom/commit/53af90d68c723f644a5a41dd273a606117109866))
+* **proxy:** bound upstream calls and hot-path costs ([#2852](https://github.com/headroomlabs-ai/headroom/issues/2852)) ([f624d3a](https://github.com/headroomlabs-ai/headroom/commit/f624d3a00ac271db7947443ddeb0c8bc2e93d3eb))
+* **subscription:** skip transcripts older than the window in compute_window_tokens ([#2861](https://github.com/headroomlabs-ai/headroom/issues/2861)) ([91d6bf3](https://github.com/headroomlabs-ai/headroom/commit/91d6bf33cde777b541375fb182d4479fdd78f81b))
+
+
+### Dependencies
+
+* bump brace-expansion from 5.0.7 to 5.0.9 in /docs ([#2751](https://github.com/headroomlabs-ai/headroom/issues/2751)) ([56ee57b](https://github.com/headroomlabs-ai/headroom/commit/56ee57be98bf109f0a46de522724ef169a4bc51c))
+* bump bytesize from 1.3.3 to 2.4.2 ([#2286](https://github.com/headroomlabs-ai/headroom/issues/2286)) ([6448545](https://github.com/headroomlabs-ai/headroom/commit/6448545a7f5a1dee88bce6f0830bdbfd1c99c617))
+* bump hf-hub from 0.4.3 to 0.5.0 ([#2285](https://github.com/headroomlabs-ai/headroom/issues/2285)) ([4925bf6](https://github.com/headroomlabs-ai/headroom/commit/4925bf6a829735977bab5000b469c3edb19c75b1))
+* bump next from 16.2.10 to 16.3.0 in /docs ([#2750](https://github.com/headroomlabs-ai/headroom/issues/2750)) ([0fd0b99](https://github.com/headroomlabs-ai/headroom/commit/0fd0b996a4b58a166491b145f4d3885c21b27cc0))
+* bump postcss from 8.5.19 to 8.5.25 in /plugins/openclaw ([#2749](https://github.com/headroomlabs-ai/headroom/issues/2749)) ([cd60ee9](https://github.com/headroomlabs-ai/headroom/commit/cd60ee9ae886b32ba5da3203e35bb6b088031fd3))
+* bump postcss from 8.5.19 to 8.5.25 in /plugins/opencode ([#2748](https://github.com/headroomlabs-ai/headroom/issues/2748)) ([ff4e016](https://github.com/headroomlabs-ai/headroom/commit/ff4e0167bbccbd4ae51bf23ddec144e61c94cd68))
+* bump postcss from 8.5.19 to 8.5.25 in /sdk/typescript ([#2747](https://github.com/headroomlabs-ai/headroom/issues/2747)) ([267c2bd](https://github.com/headroomlabs-ai/headroom/commit/267c2bdcb56e132b2dd9c065dab3498dbf730ca3))
+* bump postcss from 8.5.19 to 8.5.26 in /docs ([#2881](https://github.com/headroomlabs-ai/headroom/issues/2881)) ([e6e5826](https://github.com/headroomlabs-ai/headroom/commit/e6e5826423a0a700a8c544ce2c8cbcdef694160e))
+* bump ruff from 0.15.17 to 0.15.22 in the pip-minor-patch group ([#2501](https://github.com/headroomlabs-ai/headroom/issues/2501)) ([ecf130d](https://github.com/headroomlabs-ai/headroom/commit/ecf130d3ac6fb864098cb93fafd2621ae3ac7e12))
+* bump rusqlite from 0.32.1 to 0.40.1 ([#2287](https://github.com/headroomlabs-ai/headroom/issues/2287)) ([522faa1](https://github.com/headroomlabs-ai/headroom/commit/522faa1a59aa94e4adfd4a4afe0202d1126e187d))
+* bump the cargo-minor-patch group across 1 directory with 22 updates ([#2916](https://github.com/headroomlabs-ai/headroom/issues/2916)) ([148d860](https://github.com/headroomlabs-ai/headroom/commit/148d8605e2087f3c8d6a3fa4b8d248ad2da5858f))
+
 ## [0.34.0](https://github.com/headroomlabs-ai/headroom/compare/v0.33.0...v0.34.0) (2026-08-05)
 
 

headroom/binaries.py

@@ -29,6 +29,7 @@ import subprocess
 import sys
 import tarfile
 import tempfile
+import time
 import urllib.error
 import urllib.parse
 import urllib.request
@@ -245,12 +246,24 @@ def _download(url: str, dest: Path, *, progress: bool = True) -> None:
         raise OSError(f"binary cache directory is not writable: {dest.parent}")
     final_url = _mirror_url(url)
     req = urllib.request.Request(final_url, headers={"User-Agent": "headroom-binaries/1"})
-    try:
+    attempts = 3
-        with urllib.request.urlopen(req, timeout=60) as resp:  # noqa: S310 (https)
+    for attempt in range(1, attempts + 1):
-            total = int(resp.headers.get("Content-Length") or 0)
+        try:
-            _stream_to(resp, dest, total, label=dest.name, show_progress=progress)
+            with urllib.request.urlopen(req, timeout=60) as resp:  # noqa: S310 (https)
-    except urllib.error.URLError as e:
+                total = int(resp.headers.get("Content-Length") or 0)
-        raise BinaryFetchError(f"failed to download {final_url}: {e}") from e
+                _stream_to(resp, dest, total, label=dest.name, show_progress=progress)
+            return
+        except urllib.error.URLError as e:
+            dest.unlink(missing_ok=True)
+            if attempt == attempts:
+                raise BinaryFetchError(
+                    f"failed to download {final_url} after {attempts} attempts: {e}"
+                ) from e
+            # GitHub release assets occasionally return a transient 5xx or
+            # reset while redirecting to the object store. A short bounded
+            # retry keeps proxy startup reliable without hiding persistent
+            # credential, mirror, or connectivity failures.
+            time.sleep(0.25 * attempt)
 
 
 def _stream_to(src: Any, dest: Path, total: int, *, label: str, show_progress: bool) -> None:

plugins/headroom-agent-hooks/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "headroom",
-  "version": "0.34.0",
+  "version": "0.35.0",
   "description": "Headroom startup hooks for Claude Code and GitHub Copilot CLI.",
   "author": {
     "name": "Headroom Contributors",

plugins/headroom-agent-hooks/.github/plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "headroom",
-  "version": "0.34.0",
+  "version": "0.35.0",
   "description": "Headroom startup hooks for Claude Code and GitHub Copilot CLI.",
   "author": {
     "name": "Headroom Contributors",

plugins/openclaw/package.json

@@ -1,6 +1,6 @@
 {
   "name": "headroom-openclaw",
-  "version": "0.34.0",
+  "version": "0.35.0",
   "description": "Headroom context compression plugin for OpenClaw — 70-90% token savings with zero LLM calls",
   "type": "module",
   "main": "./dist/index.js",

plugins/opencode/package.json

@@ -1,6 +1,6 @@
 {
   "name": "headroom-opencode",
-  "version": "0.32.0",
+  "version": "0.35.0",
   "description": "Headroom proxy integration plugin for OpenCode - routes LLM traffic through the Headroom proxy for token compression",
   "type": "module",
   "main": "./dist/index.js",

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "maturin"
 
 [project]
 name = "headroom-ai"
-version = "0.34.0"
+version = "0.35.0"
 description = "The Context Optimization Layer for LLM Applications - Cut costs by 50-90%"
 readme = "README.md"
 license = "Apache-2.0"

sdk/typescript/package.json

@@ -1,6 +1,6 @@
 {
   "name": "headroom-ai",
-  "version": "0.34.0",
+  "version": "0.35.0",
   "description": "Compress LLM context. Save tokens. Fit more into every request.",
   "type": "module",
   "main": "./dist/index.cjs",

server.json

@@ -9,13 +9,13 @@
     "source": "github",
     "id": "1129940957"
   },
-  "version": "0.34.0",
+  "version": "0.35.0",
   "packages": [
     {
       "registryType": "pypi",
       "registryBaseUrl": "https://pypi.org",
       "identifier": "headroom-ai",
-      "version": "0.34.0",
+      "version": "0.35.0",
       "runtimeHint": "uvx",
       "runtimeArguments": [
         {

tests/test_binaries.py

@@ -257,6 +257,30 @@ def test_fetch_extract_zip(monkeypatch, fake_urlopen):
     assert path.read_bytes() == payload
 
 
+def test_download_retries_transient_network_failure(monkeypatch, tmp_path):
+    attempts = 0
+    sleeps: list[float] = []
+
+    def flaky_urlopen(req, timeout=None):  # noqa: ARG001
+        nonlocal attempts
+        attempts += 1
+        if attempts < 3:
+            import urllib.error
+
+            raise urllib.error.URLError("temporary GitHub release failure")
+        return _FakeResponse(b"binary")
+
+    monkeypatch.setattr(binaries.urllib.request, "urlopen", flaky_urlopen)
+    monkeypatch.setattr(binaries.time, "sleep", sleeps.append)
+
+    destination = tmp_path / "download"
+    binaries._download("https://github.com/example/tool", destination, progress=False)
+
+    assert attempts == 3
+    assert sleeps == [0.25, 0.5]
+    assert destination.read_bytes() == b"binary"
+
+
 def test_sha256_mismatch_raises_and_deletes(monkeypatch, fake_urlopen, tmp_path):
     _set_platform(monkeypatch, sys_plat="darwin", machine="arm64")
     monkeypatch.setattr(binaries.shutil, "which", lambda _name: None)

tests/test_proxy_debug_endpoints.py

@@ -35,7 +35,11 @@ from headroom.proxy.ws_session_registry import (
 
 
 @pytest.fixture
-def client():
+def client(monkeypatch):
+    # Debug endpoint tests must not depend on live upstream network access.
+    # Dedicated health-check tests cover both successful and failed upstream
+    # probes in tests/test_proxy_healthchecks.py.
+    monkeypatch.setenv("HEADROOM_SKIP_UPSTREAM_CHECK", "1")
     config = ProxyConfig(
         optimize=False,
         cache_enabled=False,