This release includes breaking changes for platform teams planning a safe upgrade.
✓ No known CVEs patched in this version
Topics
+4 more
Summary
AI summaryMulti-agent support enables a single engram instance to serve multiple agents concurrently via namespace isolation.
Full changelog
What's New
Persistent Embed Cache
Query embeddings are now cached in SQLite (128-entry FIFO). Survives restarts — repeated queries skip the embedding API entirely.
API Request Logging
Every API call is logged with method, path, namespace, and elapsed time. Finally visible who's calling what and how long it takes.
Multi-Agent Support
A single engram instance safely serves multiple agents concurrently. SQLite WAL mode + connection pool + RwLock vector index. Use X-Namespace header to isolate each agent's memory space.
Performance
- Naming LLM calls reduced ~90%: Topics track member count at naming time, only re-name when membership grows 50%+.
- bytemuck zero-copy embedding conversion
- backon structured retry replacing hand-rolled loops
Refactoring
- Vector math (
cosine_similarity,mean_vector,l2_normalize) unified into singleutil.rs— removed ~50 lines of duplication - Safe CJK-aware string slicing via
util::short_id()
Tests
- 11 new vector util tests
- 136-line scoring formula test suite
- All 280+ tests passing, zero clippy warnings
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 kael-bit/engram-rs
Hierarchical memory engine for AI agents with automatic decay, promotion, semantic dedup, and self-organizing topic tree. Single Rust binary, zero external dependencies.
Related context
Related tools
Beta — feedback welcome: [email protected]