Skip to content

This release adds 3 notable features for engineering teams evaluating rollout.

✓ No known CVEs patched
Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

agent-framework agent-orchestration ai-infra ai-interface context-engineering context-management
+4 more
go long-horizon-agents long-horizon-intelligence rust

Summary

AI summary

Explicit-source reads now fail-open by default, and Go/Rust cutover enhancements dramatically reduce tail latency.

Full changelog

ContextLattice v3.3.0

This release includes all updates merged since v3.2.16.

Commit Delta Since v3.2.16

  • c417d22 — fix(public): keep explicit source reads fail-open and document async access pattern (#119)
  • 718908e — sync(public): port agreed issue96 Go/Rust and memory-bank enhancements (#120)

What Changed

1) Retrieval UX integrity: explicit-source fail-open

From #119:

  • Explicit source requests on /memory/search now remain fail-open by default (unless blocking=true or sync_slow_sources=true).
  • Staged fetch behavior is preserved for explicit source selection.
  • Successful responses now consistently return continuation metadata instead of hard transport timeout failure patterns.
  • Access-pattern docs were updated to include /v1/memory/neighbors and staged async guidance.

Files updated in #119:

  • services/gateway-go/main.go
  • services/gateway-go/main_test.go
  • README.md
  • docs/human_agent_instruction_playbook.md
  • docs/public_overview/templates/AGENTS.contextlattice.md

2) Public Go/Rust cutover tranche + memory-bank upgrades

From #120:

  • Ported agreed dev/paid enhancements into public free runtime.
  • Expanded Go-owned runtime surfaces in gateway-go so critical memory/search paths are Go-native.
  • Added memory-bank B2 preset controls and hedge wiring.
  • Added C2 duplicate-history reduction/compaction logic in gateway memory store.
  • Added/updated phase-B soak monitor scripts and hardening for long-running launcher behavior.
  • Fixed benchmark attribution under strict Go runtime.

Primary runtime/config updates from #120:

  • services/gateway-go/main.go
  • services/gateway-go/main_test.go
  • services/gateway-go/memory_async_status.go
  • services/gateway-go/memory_browser_context.go
  • services/gateway-go/memory_context_pack.go
  • services/gateway-go/memory_misc_routes.go
  • services/gateway-go/memory_profiles.go
  • services/gateway-go/memory_store.go
  • services/gateway-go/memory_store_test.go
  • services/gateway-go/memory_v1.go
  • services/gateway-go/ops_endpoints.go
  • services/gateway-go/source_native_adapters.go
  • services/gateway-go/source_native_adapters_test.go
  • services/gateway-go/write_ingress.go
  • services/gateway-go/write_policy.go
  • services/gateway-go/go.mod
  • services/gateway-go/go.sum
  • config/env/strict_runtime.env
  • docker-compose.yml
  • docker-compose.lite.yml
  • .env.example
  • bench/backend_lane_matrix.py
  • scripts/retrieval_soak_monitor.py
  • scripts/retrieval_soak_summary.py
  • scripts/start_phase_b_step5_soak.sh
  • scripts/stop_phase_b_step5_soak.sh
  • README.md

Validation Summary

  • cd services/gateway-go && go test ./...
  • bash -n scripts/start_phase_b_step5_soak.sh
  • python3 -m py_compile scripts/retrieval_soak_monitor.py scripts/retrieval_soak_summary.py

Operational Outcome

  • Public lane now includes the agreed Go/Rust cutover enhancements and memory-bank controls from the paid/dev track subset selected for public stability and integrity.
  • Async/staged retrieval behavior for explicit-source reads is more robust and user-safe under load.

Performance Delta vs v3.2.16

Benchmark scope:

  • Workload: explicit-source retrieval (topic_rollups,memory_bank,letta)
  • Modes: balanced,deep
  • Base URL: http://127.0.0.1:8075
  • Soak duration: ~72s per release run
  • Harness: scripts/retrieval_soak_monitor.py + scripts/retrieval_soak_summary.py
  • Baseline: v3.2.16 (505bf94), Current: v3.3.0 (718908e)

| Metric | v3.2.16 | v3.3.0 | Delta | Speedup |
|---|---:|---:|---:|---:|
| balanced p95 (ms) | 9611.589 | 70.029 | -9541.560 | 137.25x |
| balanced p99 (ms) | 13739.844 | 92.044 | -13647.800 | 149.28x |
| deep p95 (ms) | 10063.372 | 45.582 | -10017.790 | 220.77x |
| deep p99 (ms) | 14386.592 | 51.450 | -14335.142 | 279.62x |
| overall degraded rate | 0.125 | 0.000 | -0.125 | n/a |

Interpretation:

  • Tail latency improved substantially in this retrieval path.
  • Degraded responses in this test path dropped from 12.5% to 0%.
  • This aligns with the explicit-source fail-open and Go/Rust runtime cutover tranche included in this release.

Repro artifacts (local benchmark run):

  • reports/release_perf_v3_2_16_vs_v3_3_0_20260331T222601Z/v3_2_16.ndjson
  • reports/release_perf_v3_2_16_vs_v3_3_0_20260331T222601Z/v3_3_0.ndjson
  • reports/release_perf_v3_2_16_vs_v3_3_0_20260331T222601Z/v3_2_16.summary.json
  • reports/release_perf_v3_2_16_vs_v3_3_0_20260331T222601Z/v3_3_0_vs_v3_2_16.summary.json

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

Track sheawinkler/ContextLattice

Get notified when new releases ship.

Sign up free

About sheawinkler/ContextLattice

Private-by-default memory and context layer for agents with Go/Rust runtime, staged retrieval across fused data backends, and long-horizon context continuity.

All releases →

Beta — feedback welcome: [email protected]