[D] antaris-suite 3.0 (open source, free) — zero-dependency agent memory, guard, routing, and context management (benchmarks + 3-model code review inside)
Open-source Python suite eliminates API dependencies for agent memory, routing, and safety with 1.9ms search latency.
The Antaris team has launched antaris-suite 3.0, a comprehensive open-source infrastructure toolkit for building and managing AI agents. The suite consists of six zero-dependency Python packages: antaris-memory (BM25 search with JSONL storage), antaris-guard (configurable policy engine), antaris-router (cost-aware routing), antaris-context (token budget management), antaris-pipeline (agent turn coordination), and antaris-contracts. Benchmarks on an M4 Mac Mini show 1.9ms search latency and 11,600 items/second bulk ingest performance, outperforming alternatives like mem0. The system uses portable JSONL storage with write-ahead logging for crash safety and integrates natively with OpenClaw via plugins. Unlike vector database-dependent solutions, it runs entirely in-process with no external API requirements.
- Zero-dependency Python packages for agent memory, routing, safety, and context management with 1.9ms search latency
- BM25-based memory system with JSONL storage handles 11,600 items/sec ingest and includes crash-safe WAL logging
- Native OpenClaw plugin integration enables automatic memory recall/ingest without code changes for existing users
Why It Matters
Eliminates vendor lock-in and API costs for agent infrastructure while providing enterprise-grade performance for production AI systems.