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


<details><summary>0.29.0</summary>

##
[0.29.0](v0.28.0...v0.29.0)
(2026-07-03)


### Features

* **proxy:** add --lossless no-CCR mode with format-native compaction
([#1721](#1721))
([c75ebde](c75ebde))
* **stats:** surface Codex WS compression counters in /stats summary
([#1680](#1680))
([2fe19c3](2fe19c3))
* **transforms:** adaptive Otsu KEEP/DROP threshold (+ land relevance
split on main)
([#1726](#1726))
([eea667a](eea667a))


### Bug Fixes

* **bedrock:** fail fast when session-token auth lacks botocore
([#1553](#1553))
([54cfa36](54cfa36))
* **bedrock:** route ARNs via converse, named AWS profiles, and au. re…
([#1456](#1456))
([7d87aa2](7d87aa2))
* **ccr:** honor workspace dir for sqlite store
([#1564](#1564))
([96e1dfe](96e1dfe))
* **claude:** surface Remote Control proxy incompatibility
([#1610](#1610))
([4bf7f92](4bf7f92))
* **cli:** stop advertising unwired compression tuning env vars in
banner
([#1634](#1634))
([d5bf98d](d5bf98d))
* **codex:** avoid duplicate headroom provider config
([#1431](#1431))
([ddd4adf](ddd4adf))
* **compression:** reject lossy unmarked tool output in unit router path
([#1479](#1479))
([de24cd5](de24cd5))
* **cortex-code:** migrate to current Cortex REST API endpoints + add
e2e benchmarks
([#1474](#1474))
([f00ace6](f00ace6))
* **dashboard:** align token savings headline denominator
([#1653](#1653))
([646e705](646e705))
* **dashboard:** derive per-project setup URL from live origin
([#1511](#1511))
([e035aef](e035aef))
* **detection:** contain unidiff panic on orphaned +++ target line
([#1548](#1548))
([e386c09](e386c09))
* **evals:** CJK-aware F1 tokenization + token estimation
([#1527](#1527))
([99a8540](99a8540))
* **install:** close parent log fd in start_detached_agent
([#1576](#1576))
([816cb85](816cb85))
* **install:** use Windows-safe PID liveness probe in runtime_status
([#1544](#1544))
([#1560](#1560))
([6b227b9](6b227b9))
* **learn:** aggregate verbosity baselines across projects instead of
overwriting
([#1288](#1288))
([27a5468](27a5468))
* **mcp:** show lifetime totals and label rolling session scope in
headroom_stats
([#1428](#1428))
([1c0e152](1c0e152))
* **memory:** cap local embedder CPU thread oversubscription
([#198](#198))
([#1559](#1559))
([b84afbf](b84afbf))
* **memory:** singleflight LocalBackend init to stop cold-start races
([#1691](#1691))
([bec47a1](bec47a1))
* **openclaw:** detect uv-installed headroom binary in ~/.local/bin
([#1459](#1459))
([adaeb88](adaeb88))
* **opencode:** preserve custom OpenAI gateway paths
([#1596](#1596))
([c19347c](c19347c))
* **opencode:** route native providers + load transport plugin, fix
Serena context
([#1573](#1573))
([ad0034f](ad0034f))
* preserve anthropic passthrough tool order
([#1427](#1427))
([a932247](a932247))
* **proxy/auth:** match real Anthropic OAuth token prefix (sk-ant-oat)
([#1672](#1672))
([8cddf9b](8cddf9b))
* **proxy:** expose persistent savings metrics
([#1647](#1647))
([5fe4e7b](5fe4e7b))
* **proxy:** fail open when kompress saturation would exhaust
pre-upstream budget
([#1430](#1430))
([15ac650](15ac650))
* **proxy:** handle streaming CCR retrieval
([#1451](#1451))
([d337e3b](d337e3b))
* **proxy:** include system/tools/sampling in cache key
([#1473](#1473))
([312129a](312129a))
* **proxy:** preserve Responses passthrough bytes
([#1598](#1598))
([2a34a82](2a34a82))
* **proxy:** strip Codex lite header on the HTTP /responses path
([#1663](#1663))
([9fbd47b](9fbd47b))
* **proxy:** wire --compression-max-workers /
HEADROOM_COMPRESSION_MAX_WORKERS
([#1632](#1632))
([814ffa3](814ffa3))
* **savings:** count cache-read tokens in input cost estimate
([#1429](#1429))
([72ade37](72ade37))
* skip Magika backend on x86 CPUs without AVX2
([#1162](#1162))
([64783d8](64783d8))
* **transforms/content-router:** route grep/log output away from HTML
extractor
([#1719](#1719))
([0d18ef2](0d18ef2))
* **transforms:** bound native content detection with a Windows watchdog
([#575](#575))
([#1563](#1563))
([95abca3](95abca3))
* Vertex AI support for Claude Code with ANTHROPIC_VERTEX_BASE_URL
([#1393](#1393))
([cff7247](cff7247))
* **wrap:** detach the shared proxy on Windows so it survives an
ungraceful agent close
([#1464](#1464))
([6cba441](6cba441))
* **wrap:** preserve custom Vertex base URL
([#1477](#1477))
([75427bb](75427bb))
* **wrap:** remove rtk instructions from Codex AGENTS.md on unwrap
([#1604](#1604))
([c9d717c](c9d717c))
</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: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

…labs-ai#198) (headroomlabs-ai#1559)

## Description

The torch/sentence-transformers `LocalEmbedder` ran encodes on the
shared default executor with **no BLAS/OpenMP thread cap**. Under
concurrent load each `encode()` fans out to ~`os.cpu_count()`
BLAS/OpenMP threads, so N in-flight encodes spawn ~`N × cpu_count` OS
threads — oversubscribing the CPU, slowing the `memory_context` stage
and (on smaller boxes) starving the asyncio event loop. The ONNX
embedder already bounds its threads
(`create_cpu_session_options(intra_op_num_threads=1,
inter_op_num_threads=1)`); this brings the torch path to parity.

Supersedes headroomlabs-ai#691 by @oxura — closed only for the open-PR cap, with an
explicit invitation to resubmit; no technical objection was raised, and
its CI was fully green. Credit to @oxura for the original diagnosis and
fix. That PR capped threads by setting BLAS/OpenMP env vars at import
time plus `torch.set_num_threads`; this PR instead runs CPU encodes on a
dedicated, size-limited executor whose workers each pin their thread
pool — which additionally bounds in-flight encode concurrency (the
issue's Fix B/C) and keeps the cap contained to the embedder rather than
mutating process-global env at import.

Closes headroomlabs-ai#198

## Type of Change

- [x] Bug fix (non-breaking change that fixes an issue)
- [x] Performance improvement

## Changes Made

- CPU encodes now run on a **dedicated, size-limited executor** whose
worker `initializer` pins each worker's torch intra-op pool (and sets
BLAS/OpenMP env defaults). torch's OpenMP thread count is per-thread, so
a one-shot cap misses pooled executor workers — the per-worker
initializer caps every worker deterministically.
- Total embedding threads are bounded by `HEADROOM_EMBED_CONCURRENCY`
(default `min(4, os.cpu_count())`) × `HEADROOM_EMBED_NUM_THREADS`
(default `1`); invalid/non-positive values fall back safely (≥1).
- Mirrors the existing MPS dedicated-single-worker-executor pattern;
CUDA keeps the shared default executor (GPU compute is off-CPU).
`setdefault` never overrides an operator's explicit `OMP_NUM_THREADS`.

## Testing

- [x] Unit tests pass (`pytest`)
- [x] Linting passes (`ruff check .`)
- [x] Type checking passes (`mypy headroom`)
- [x] New tests added for new functionality
- [x] Manual testing performed

### Test Output

```text
$ uv run pytest tests/test_memory/test_embedder_thread_cap.py tests/test_memory/test_embedder_mps_serialization.py -q
13 passed

$ uv run pytest tests/test_memory/ tests/test_cli_proxy_embedding_server.py -q
533 passed        # no regressions from the executor change

$ uv run ruff check .  &&  uv run ruff format --check .
All checks passed!   /   1016 files already formatted

$ uv run mypy headroom --ignore-missing-imports
Success: no issues found in 404 source files
```

New `tests/test_memory/test_embedder_thread_cap.py`: env resolution for
both knobs (default / positive / invalid / clamped), worker-init env
application + operator-override safety, and a behavioral test that loads
the real CPU embedder and asserts every executor worker is pinned to the
configured intra-op thread count. Updated
`test_embedder_mps_serialization.py` to the new CPU contract.

## Real Behavior Proof

- Environment: built this branch into a CPU-only Linux container,
removed `onnxruntime` so the proxy falls back to the torch
`LocalEmbedder`; a container has no MPS/CUDA, so it resolves to
`device=cpu` — the deployment where headroomlabs-ai#198 occurs. Python 3.12, torch
2.12.1, `all-MiniLM-L6-v2`, container capped to 4 CPUs, 32 concurrent
clients.
- Exact command / steps: `headroom proxy --host 0.0.0.0 --memory`
in-container; a concurrent `/v1/messages` driver from the host (invalid
key — `memory_context` runs before the upstream call); measured the
`memory_context` stage from `/metrics` before vs after the cap.
- Observed result: the embedder stage this PR targets improved —
`memory_context` avg 73.5 ms → 58.7 ms and max 279 ms → 242 ms (uncapped
12×8 = 96 threads vs fix 4×1): ~20% faster and steadier inside the real
proxy. Isolated component benchmarks (heavy concurrent `embed_batch`;
`LocalBackend.search_memories`) show a larger effect — tail event-loop
stall ~16–24 ms → ~3 ms, and search throughput +57%. Unit/regression: 13
new tests + 533 memory-suite tests pass; `ruff` + `mypy` clean.
- Not tested: the issue's absolute multi-second `/livez` spike. On my
hardware/synthetic load, `/livez` stalls were dominated by the
upstream-connection path (invalid-key DNS/TLS), not the ~250 ms
`memory_context` stage, so I can't attribute the multi-second figure to
the embedder here — the original report was on an 8-core box with real
Claude Code transcripts that drove `memory_context` itself to several
seconds. Linux/CUDA hardware not exercised; no live LLM provider used;
ONNX path unchanged. This PR removes the documented thread
oversubscription and brings the torch path to ONNX parity; it does not
claim to single-handedly resolve the 4 s figure.

Measured `memory_context` stage timing (real containerized proxy, torch
CPU embedder, 4 CPUs, 32 concurrent clients):

| `memory_context` | avg | max |
|---|---|---|
| Before (uncapped, 12×8 = 96 threads) | 73.5 ms | 279 ms |
| After (fix, 4×1) | 58.7 ms | 242 ms |

## Review Readiness

- [x] I have performed a self-review
- [x] This PR is ready for human review

## Additional Notes

Default-behavior change: CPU encodes use a dedicated bounded pool
instead of the shared default executor (`close()` tears it down). Both
knobs are opt-in overrides with safe defaults. No new dependencies.

Signed-off-by: Krishnachaitanyakc <krishnabkc15@gmail.com>

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


<details><summary>0.29.0</summary>

##
[0.29.0](headroomlabs-ai/headroom@v0.28.0...v0.29.0)
(2026-07-03)


### Features

* **proxy:** add --lossless no-CCR mode with format-native compaction
([headroomlabs-ai#1721](headroomlabs-ai#1721))
([c75ebde](headroomlabs-ai@c75ebde))
* **stats:** surface Codex WS compression counters in /stats summary
([headroomlabs-ai#1680](headroomlabs-ai#1680))
([2fe19c3](headroomlabs-ai@2fe19c3))
* **transforms:** adaptive Otsu KEEP/DROP threshold (+ land relevance
split on main)
([headroomlabs-ai#1726](headroomlabs-ai#1726))
([eea667a](headroomlabs-ai@eea667a))


### Bug Fixes

* **bedrock:** fail fast when session-token auth lacks botocore
([headroomlabs-ai#1553](headroomlabs-ai#1553))
([54cfa36](headroomlabs-ai@54cfa36))
* **bedrock:** route ARNs via converse, named AWS profiles, and au. re…
([headroomlabs-ai#1456](headroomlabs-ai#1456))
([7d87aa2](headroomlabs-ai@7d87aa2))
* **ccr:** honor workspace dir for sqlite store
([headroomlabs-ai#1564](headroomlabs-ai#1564))
([96e1dfe](headroomlabs-ai@96e1dfe))
* **claude:** surface Remote Control proxy incompatibility
([headroomlabs-ai#1610](headroomlabs-ai#1610))
([4bf7f92](headroomlabs-ai@4bf7f92))
* **cli:** stop advertising unwired compression tuning env vars in
banner
([headroomlabs-ai#1634](headroomlabs-ai#1634))
([d5bf98d](headroomlabs-ai@d5bf98d))
* **codex:** avoid duplicate headroom provider config
([headroomlabs-ai#1431](headroomlabs-ai#1431))
([ddd4adf](headroomlabs-ai@ddd4adf))
* **compression:** reject lossy unmarked tool output in unit router path
([headroomlabs-ai#1479](headroomlabs-ai#1479))
([de24cd5](headroomlabs-ai@de24cd5))
* **cortex-code:** migrate to current Cortex REST API endpoints + add
e2e benchmarks
([headroomlabs-ai#1474](headroomlabs-ai#1474))
([f00ace6](headroomlabs-ai@f00ace6))
* **dashboard:** align token savings headline denominator
([headroomlabs-ai#1653](headroomlabs-ai#1653))
([646e705](headroomlabs-ai@646e705))
* **dashboard:** derive per-project setup URL from live origin
([headroomlabs-ai#1511](headroomlabs-ai#1511))
([e035aef](headroomlabs-ai@e035aef))
* **detection:** contain unidiff panic on orphaned +++ target line
([headroomlabs-ai#1548](headroomlabs-ai#1548))
([e386c09](headroomlabs-ai@e386c09))
* **evals:** CJK-aware F1 tokenization + token estimation
([headroomlabs-ai#1527](headroomlabs-ai#1527))
([99a8540](headroomlabs-ai@99a8540))
* **install:** close parent log fd in start_detached_agent
([headroomlabs-ai#1576](headroomlabs-ai#1576))
([816cb85](headroomlabs-ai@816cb85))
* **install:** use Windows-safe PID liveness probe in runtime_status
([headroomlabs-ai#1544](headroomlabs-ai#1544))
([headroomlabs-ai#1560](headroomlabs-ai#1560))
([6b227b9](headroomlabs-ai@6b227b9))
* **learn:** aggregate verbosity baselines across projects instead of
overwriting
([headroomlabs-ai#1288](headroomlabs-ai#1288))
([27a5468](headroomlabs-ai@27a5468))
* **mcp:** show lifetime totals and label rolling session scope in
headroom_stats
([headroomlabs-ai#1428](headroomlabs-ai#1428))
([1c0e152](headroomlabs-ai@1c0e152))
* **memory:** cap local embedder CPU thread oversubscription
([headroomlabs-ai#198](headroomlabs-ai#198))
([headroomlabs-ai#1559](headroomlabs-ai#1559))
([b84afbf](headroomlabs-ai@b84afbf))
* **memory:** singleflight LocalBackend init to stop cold-start races
([headroomlabs-ai#1691](headroomlabs-ai#1691))
([bec47a1](headroomlabs-ai@bec47a1))
* **openclaw:** detect uv-installed headroom binary in ~/.local/bin
([headroomlabs-ai#1459](headroomlabs-ai#1459))
([adaeb88](headroomlabs-ai@adaeb88))
* **opencode:** preserve custom OpenAI gateway paths
([headroomlabs-ai#1596](headroomlabs-ai#1596))
([c19347c](headroomlabs-ai@c19347c))
* **opencode:** route native providers + load transport plugin, fix
Serena context
([headroomlabs-ai#1573](headroomlabs-ai#1573))
([ad0034f](headroomlabs-ai@ad0034f))
* preserve anthropic passthrough tool order
([headroomlabs-ai#1427](headroomlabs-ai#1427))
([a932247](headroomlabs-ai@a932247))
* **proxy/auth:** match real Anthropic OAuth token prefix (sk-ant-oat)
([headroomlabs-ai#1672](headroomlabs-ai#1672))
([8cddf9b](headroomlabs-ai@8cddf9b))
* **proxy:** expose persistent savings metrics
([headroomlabs-ai#1647](headroomlabs-ai#1647))
([5fe4e7b](headroomlabs-ai@5fe4e7b))
* **proxy:** fail open when kompress saturation would exhaust
pre-upstream budget
([headroomlabs-ai#1430](headroomlabs-ai#1430))
([15ac650](headroomlabs-ai@15ac650))
* **proxy:** handle streaming CCR retrieval
([headroomlabs-ai#1451](headroomlabs-ai#1451))
([d337e3b](headroomlabs-ai@d337e3b))
* **proxy:** include system/tools/sampling in cache key
([headroomlabs-ai#1473](headroomlabs-ai#1473))
([312129a](headroomlabs-ai@312129a))
* **proxy:** preserve Responses passthrough bytes
([headroomlabs-ai#1598](headroomlabs-ai#1598))
([2a34a82](headroomlabs-ai@2a34a82))
* **proxy:** strip Codex lite header on the HTTP /responses path
([headroomlabs-ai#1663](headroomlabs-ai#1663))
([9fbd47b](headroomlabs-ai@9fbd47b))
* **proxy:** wire --compression-max-workers /
HEADROOM_COMPRESSION_MAX_WORKERS
([headroomlabs-ai#1632](headroomlabs-ai#1632))
([814ffa3](headroomlabs-ai@814ffa3))
* **savings:** count cache-read tokens in input cost estimate
([headroomlabs-ai#1429](headroomlabs-ai#1429))
([72ade37](headroomlabs-ai@72ade37))
* skip Magika backend on x86 CPUs without AVX2
([headroomlabs-ai#1162](headroomlabs-ai#1162))
([64783d8](headroomlabs-ai@64783d8))
* **transforms/content-router:** route grep/log output away from HTML
extractor
([headroomlabs-ai#1719](headroomlabs-ai#1719))
([0d18ef2](headroomlabs-ai@0d18ef2))
* **transforms:** bound native content detection with a Windows watchdog
([headroomlabs-ai#575](headroomlabs-ai#575))
([headroomlabs-ai#1563](headroomlabs-ai#1563))
([95abca3](headroomlabs-ai@95abca3))
* Vertex AI support for Claude Code with ANTHROPIC_VERTEX_BASE_URL
([headroomlabs-ai#1393](headroomlabs-ai#1393))
([cff7247](headroomlabs-ai@cff7247))
* **wrap:** detach the shared proxy on Windows so it survives an
ungraceful agent close
([headroomlabs-ai#1464](headroomlabs-ai#1464))
([6cba441](headroomlabs-ai@6cba441))
* **wrap:** preserve custom Vertex base URL
([headroomlabs-ai#1477](headroomlabs-ai#1477))
([75427bb](headroomlabs-ai@75427bb))
* **wrap:** remove rtk instructions from Codex AGENTS.md on unwrap
([headroomlabs-ai#1604](headroomlabs-ai#1604))
([c9d717c](headroomlabs-ai@c9d717c))
</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: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>