About Suprflo

Suprflo is a production-grade memory layer designed for AI agents. It addresses the stateless nature of LLMs by providing a robust system for extracting, storing, and recalling information from conversations. This ensures that AI agents can maintain context, remember user preferences, and compound knowledge over time, overcoming the limitations of short context windows.

Key Features:

• Bi-temporal Truth: Facts have validity windows, meaning they are never overwritten. You can query the state of knowledge as it existed on any past date, providing a complete audit trail.

• Recall Floor: Every raw turn of a conversation is preserved alongside extracted facts, ensuring that information is never silently lost.

• Benchmarked Accuracy: Suprflo's performance is rigorously measured and published on an open benchmark harness, with every failure root-caused and audited against source conversations. It boasts 92.1% LoCoMo accuracy.

• Postgres-Native: Built on plain PostgreSQL with pgvector, offering a familiar and powerful database foundation.

• Hybrid Retrieval: Recalls information using a combination of semantic search, BM25, entity boosting, and optional cross-encoder reranking for high precision.

• Memory Differentiation: Supports semantic, episodic, and procedural memory types, along with Ebbinghaus decay and a reflection engine.

• Multi-Tenant Native: Enforces data isolation at the SQL WHERE clause level for organizations, projects, users, and agents.

• Flexible Integration: Offers a single REST API and first-party SDKs for Python and Node.js, with drop-in support for popular frameworks like LangChain, LangGraph, and LlamaIndex.

• Provider Agnostic: Abstracts LLMs, embedders, and vector stores through a factory pattern, allowing easy swapping of backends without code rewrites.

Suprflo is designed to work seamlessly with your existing AI stack, providing a reliable and scalable memory solution that empowers AI agents to be more effective and context-aware.