Problem solved

LLMs are stateless and lose user-specific facts when context ends. Suprflo solves this by extracting, storing, and retrieving facts from conversations with bi-temporal structure, allowing agents to remember preferences, constraints, and history without stuffing full transcripts into prompts.

Use cases

• Remembering user preferences and allergies across chatbot sessions
• Personalizing AI assistant responses based on past conversations
• Answering questions about what a user believed or stated at a past date
• Traversing knowledge graphs for multi-hop relational queries
• Providing auditable memory for production AI agents with full history
• Reducing prompt cost by retrieving relevant memories instead of full context
• Building long-term personalization for customer support or copilot agents

Audience

• Developers
• ML Engineers
• Product Managers
• Enterprise
• Startups

Pros

• Bi-temporal data model preserves history and supports as-of queries
• Postgres-native with no vendor lock-in
• Open benchmark harness with reproducible results
• Hybrid retrieval with semantic reranking
• Raw turns retained as a recall floor

Cons

• Not yet generally available; requires joining waitlist
• Benchmark coverage still preliminary for some datasets, such as LongMemEval
• Potential LLM extraction cost and latency overhead
• Requires integration work to fit existing agent pipelines
• Pricing model beyond free tier not detailed