Bump to v0.5.19: cache stats fixes, Codex WS fix, compression-vs-cache tracking

Token-level cache hit rate, compression-vs-cache tracking, dashboard SQL, security plan

Fix Codex WebSocket HTTP 500: ChatGPT auth routing, correct beta header, HTTP fallback (#71)

Fix remaining review issues: null safety, compressed count, batch tokens

Disable image compression until token counting is accurate

Fix code review issues: batch NameError, null safety, token types

Refactor: split server.py into handler mixins (Steps 5-9)

Refactor: extract 7 modules from server.py (Steps 2-4)

Apply Ruff formatting to rebased stack

Polish rebased display-session stack

Format display-session tracker changes

Add canonical display-session metrics for downstream dashboards

feat: add spend and input-token rollups to stats-history

Fix WebSocket SSL on Windows: use native websockets SSL handling

Disable CacheAligner: it inflates tokens and breaks prefix caching

Fix Codex WS proxy (Issue #86), LiteLLM metrics, and backend forwarding

Refactor: extract data models to headroom/proxy/models.py (Step 1/9)

Add transforms_summary to TransformResult and /v1/compress response

Merge pull request #81 from dorphalsig/main

Fix multi-worker beacon spam, cost dashboard, upsert telemetry; bump 0.5.17

Fix dashboard cost savings: use model list price, not moving average

Merge pull request #61 from gglucass/feat/persist-savings-history

Fix WebSocket proxy: forward all headers + required OpenAI-Beta header

Add Anthropic url overrides via ANTHROPIC_TARGET_URL env var and --anthropic-api-url flag

Merge remote-tracking branch 'origin/main' into feat/persist-savings-history

feat: add historical rollups and exports

Add WebSocket proxy for /v1/responses (Codex gpt-5.4+)

Add compression for OpenAI Responses API (/v1/responses)

Fix zstd streaming decompression, add zstandard dep; bump to 0.5.14

Fix streaming tool calls, compressed request bodies, beacon field names; bump to 0.5.13

Fix cost calculation: don't mix cache savings into Headroom savings

Merge branch 'main' into feat/persist-savings-history

Update savings_tracker.py

fix: restore anthropic compression when hooks enabled

Merge pull request #60 from gglucass/feat/add-headroom-sdk

OpenClaw plugin fixes, telemetry fix, cost tracker improvement, tokenBudget support

style: fix ruff format in server.py

fix: move extract_user_query import out of conditional scope

feat: persist proxy savings history

feat: allow overriding proxy telemetry sdk via HEADROOM_SDK

Add TypeScript SDK (headroom-ai npm package)

Remove LLMLingua: Kompress is the sole text compressor

Pass user query context through proxy pipeline (9 call sites)

fix: forward upstream ratelimit headers in streaming responses

Reduce compression latency: cache serializations, eager-load all compressors, fix Magika, bump to 0.5.5

Add telemetry beacon, fix Starlette 0.41+ crash, bump to 0.5.4

fix: close TOIN feedback loop — retrievals now flow back for learning

feat: add license key validation + phone-home usage reporter

fix: memory handler overwrites system prompt when it's a list of content blocks

feat: add live traffic learning + cross-agent memory writers (--learn flag)