This release includes breaking changes for platform teams planning a safe upgrade.
✓ No known CVEs patched in this version
Topics
Affected surfaces
Summary
AI summaryAdded CacheCheckOptions.judge to handle borderline cache hits with an LLM adjudicator and throws SemanticCacheUsageError when a judge is supplied.
Changes in this release
| Type | Severity | Summary | CVE |
|---|---|---|---|
| Feature | Medium |
LLM-as-judge for borderline hits added to CacheCheckOptions.judge with configurable timeoutMs and onError. LLM-as-judge for borderline hits added to CacheCheckOptions.judge with configurable timeoutMs and onError. Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Sentinel/cluster failover event tracking with slowlog correlation added to anomaly module. Sentinel/cluster failover event tracking with slowlog correlation added to anomaly module. Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Start session modal with pre-flight and 5-min confirmation added to web/monitor UI. Start session modal with pre-flight and 5-min confirmation added to web/monitor UI. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
New Prometheus metrics {prefix}_judge_decisions_total{decision} and {prefix}_judge_duration_seconds{decision}. New Prometheus metrics {prefix}_judge_decisions_total{decision} and {prefix}_judge_duration_seconds{decision}. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
New OTel span attributes cache.judge.invoked, cache.judge.decision, cache.judge.latency_ms added. New OTel span attributes cache.judge.invoked, cache.judge.decision, cache.judge.latency_ms added. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
JudgeOptions type exported from the package root. JudgeOptions type exported from the package root. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
examples/judge runnable example with real OpenAI embeddings included. examples/judge runnable example with real OpenAI embeddings included. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Automatic welcome email sent on workspace provisioning enabled. Automatic welcome email sent on workspace provisioning enabled. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Scaffold module, dev preview gate, and ping endpoint implemented in monitor. Scaffold module, dev preview gate, and ping endpoint implemented in monitor. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Capture_sessions and capture_chunks schema with list endpoint created. Capture_sessions and capture_chunks schema with list endpoint created. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
HealthGate deep module and diagnostic endpoint added to monitor. HealthGate deep module and diagnostic endpoint added to monitor. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
ProviderDetector and AclChecker with pre-flight endpoint implemented in monitor. ProviderDetector and AclChecker with pre-flight endpoint implemented in monitor. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
CaptureWriter deep module and storage chunk persistence (PR 5a of 5) added to monitor. CaptureWriter deep module and storage chunk persistence (PR 5a of 5) added to monitor. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Start/stop/get session endpoints and Valkey MONITOR wiring (PR 5b of 5) implemented. Start/stop/get session endpoints and Valkey MONITOR wiring (PR 5b of 5) implemented. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Session lifecycle webhooks (started, completed, truncated) added to monitor. Session lifecycle webhooks (started, completed, truncated) added to monitor. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
TailGateway WebSocket with pause/resume and per-viewer subscriptions introduced. TailGateway WebSocket with pause/resume and per-viewer subscriptions introduced. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Sessions list with 5s polling auto-refresh implemented in web/monitor UI. Sessions list with 5s polling auto-refresh implemented in web/monitor UI. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Live tail view with pause/resume and bounded buffer implemented in web/monitor UI. Live tail view with pause/resume and bounded buffer implemented in web/monitor UI. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Post-capture filters and JSON/CSV export added to monitor. Post-capture filters and JSON/CSV export added to monitor. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
CrossReferenceEngine and cross-reference REST endpoint (PR 12 of 25) implemented in monitor. CrossReferenceEngine and cross-reference REST endpoint (PR 12 of 25) implemented in monitor. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Cross-reference panel with baseline selector added to web/monitor UI. Cross-reference panel with baseline selector added to web/monitor UI. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Cluster per-node selector for capture sessions (PR 14a of 14) implemented in monitor. Cluster per-node selector for capture sessions (PR 14a of 14) implemented in monitor. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Cluster fan-out with partial-failure handling (PR 14b of 14) added to monitor. Cluster fan-out with partial-failure handling (PR 14b of 14) added to monitor. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Pro+ capture triggers, REST, and license gate implemented in monitor. Pro+ capture triggers, REST, and license gate implemented in monitor. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Trigger.created and session.skipped webhooks added to monitor. Trigger.created and session.skipped webhooks added to monitor. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Triggers tab UI implemented in web/monitor. Triggers tab UI implemented in web/monitor. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Capture-on-next row action with prefilled modal added to web/anomalies UI. Capture-on-next row action with prefilled modal added to web/anomalies UI. Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Low |
Start session modal includes pre‑flight check and 5‑minute confirmation countdown. Start session modal includes pre‑flight check and 5‑minute confirmation countdown. Source: granite4.1:30b@2026-05-22-audit Confidence: low |
— |
| Bugfix | Medium |
checkBatch() throws SemanticCacheUsageError when judge is supplied, consistent with rerank and staleAfterModelChange handling. checkBatch() throws SemanticCacheUsageError when judge is supplied, consistent with rerank and staleAfterModelChange handling. Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Refactor | Medium |
nearestMiss.deltaToThreshold may now be <= 0 for judge rejections. nearestMiss.deltaToThreshold may now be <= 0 for judge rejections. Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Other | Medium |
Unit tests for InferenceLatencyService added by @Vswaroop04. Unit tests for InferenceLatencyService added by @Vswaroop04. Source: llm_adapter@2026-05-21 Confidence: low |
— |
Full changelog
Added
- LLM-as-judge for borderline hits —
CacheCheckOptions.judgeaccepts ajudgeFnthat adjudicates hits whose cosine distance falls in the uncertainty band (threshold - uncertaintyBand < score <= threshold). Accepted hits are promoted toconfidence: 'high'; rejected hits are demoted to a miss withnearestMisspopulated. ConfigurabletimeoutMs(default 2000) andonError(default'accept', fail-open). - New Prometheus metrics
{prefix}_judge_decisions_total{decision}and{prefix}_judge_duration_seconds{decision}with labelsaccept | reject | error_accept | error_reject | timeout_accept | timeout_reject. - New OTel span attributes
cache.judge.invoked,cache.judge.decision,cache.judge.latency_ms. JudgeOptionstype exported from the package root.examples/judge— runnable example with real OpenAI embeddings, showing distances and matched entries.
Changed
nearestMiss.deltaToThresholdmay now be<= 0when a miss originates from a judge rejection (score cleared the threshold but the judge rejected it). Existing miss paths still produce> 0. Documented on the type.checkBatch()throwsSemanticCacheUsageErrorwhenjudgeis supplied, matching the existing handling ofrerankandstaleAfterModelChange.
Breaking changes
None.
What's Changed
- build(deps): bump uuid and @langchain/core in /packages/agent-cache/examples/langchain by @dependabot[bot] in https://github.com/BetterDB-inc/monitor/pull/140
- build(deps): bump @anthropic-ai/sdk from 0.90.0 to 0.91.1 in /packages/agent-cache/examples/anthropic by @dependabot[bot] in https://github.com/BetterDB-inc/monitor/pull/141
- fix: disable histogram polling after repeated failures by @100NikhilBro in https://github.com/BetterDB-inc/monitor/pull/143
- feat: send welcome email automatically on workspace provisioning by @KIvanow in https://github.com/BetterDB-inc/monitor/pull/180
- feat(anomaly): sentinel/cluster failover event tracking with slowlog correlation by @SBALAVIGNESH123 in https://github.com/BetterDB-inc/monitor/pull/142
- feature(monitor): scaffold module + dev preview gate + ping endpoint by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/163
- feature(monitor): capture_sessions + capture_chunks schema and list endpoint by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/164
- feature(monitor): HealthGate deep module + diagnostic endpoint by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/165
- feature(monitor): ProviderDetector + AclChecker + pre-flight endpoint by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/166
- feature(monitor): CaptureWriter deep module + storage chunk persistence (PR 5a of 5) by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/167
- feature(monitor): start/stop/get session endpoints + Valkey MONITOR wiring (PR 5b of 5) by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/168
- feature(monitor): session lifecycle webhooks (started, completed, truncated) by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/169
- feature(monitor): TailGateway WebSocket + pause/resume + per-viewer subscriptions by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/170
- feature(web/monitor): Sessions list with 5s polling auto-refresh by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/171
- feature(web/monitor): start session modal with pre-flight + 5-min confirmation by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/172
- feature(web/monitor): live tail view + pause/resume + bounded buffer by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/196
- feature(monitor): post-capture filters + JSON/CSV export by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/174
- feature(monitor): CrossReferenceEngine + cross-reference REST endpoint (PR 12 of 25) by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/175
- feature(web/monitor): cross-reference panel with baseline selector by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/176
- feature(monitor): cluster per-node selector for capture sessions (PR 14a of 14) by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/177
- feature(monitor): cluster fan-out with partial-failure handling (PR 14b of 14) by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/178
- feature(monitor): Pro+ capture triggers + REST + license gate by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/179
- feature(monitor): trigger.created + session.skipped webhooks by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/181
- feature(web/monitor): Triggers tab UI by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/182
- test(inference-latency): add unit tests for InferenceLatencyService by @Vswaroop04 in https://github.com/BetterDB-inc/monitor/pull/159
- feature(web/anomalies): Capture-on-next row action + prefilled modal by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/183
- feature(monitor): scheduled captures + CaptureScheduler + REST by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/184
- feature(monitor): Scheduled tab UI + cron expression support by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/185
- feature(monitor): capture-vs-capture diff endpoint + compare UI by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/186
- feature(monitor): data-retention pruning hooks for monitor tables by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/187
- feature(web/monitor): provider warnings + ACL snippet polish by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/188
- feature(monitor): optional value-redaction toggle for write commands by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/189
- feature(monitor): launch — remove dev preview gate, add docs + CHANGELOG by @jamby77 in https://github.com/BetterDB-inc/monitor/pull/190
- Improvements/monitor by @KIvanow in https://github.com/BetterDB-inc/monitor/pull/199
- feat(semantic-cache): LLM-as-judge for borderline hits (v0.5.0) by @KIvanow in https://github.com/BetterDB-inc/monitor/pull/193
New Contributors
- @SBALAVIGNESH123 made their first contribution in https://github.com/BetterDB-inc/monitor/pull/142
- @Vswaroop04 made their first contribution in https://github.com/BetterDB-inc/monitor/pull/159
Full Changelog: https://github.com/BetterDB-inc/monitor/compare/v0.20.0...semantic-cache-v0.5.0
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
About BetterDB-inc/monitor
Valkey-first observability with Redis compatibility. Query real-time metrics, analyze slow commands, detect hot keys, and investigate performance issues directly from AI coding assistants.
Related context
Related tools
Earlier breaking changes
- vagent-v1.5.0 TLS is now mandatory; agent exits with error if `--valkey-tls true` is not set for IAM auth mode.
Beta — feedback welcome: [email protected]