xmemory provides a structured memory layer for AI agents that converts natural‑language read and write commands into typed objects, relations, and timestamps, ensuring deduplicated, versioned, and queryable state. It offers built‑in observability, fine‑grained access controls, and token‑efficient storage, and can be deployed as a managed SaaS, zero‑retention SaaS linked to a customer database, or self‑hosted solution.
Funding
Funding not disclosed
Founders
Product
Problem
AI agents often rely on unstructured text or simple vector stores for memory, leading to duplicated, drifting, or ambiguous facts that must be reinterpreted on every read. This makes reliable state tracking, exact lookups, and relational queries difficult for production systems.
Solution
xmemory offers a memory layer that accepts natural‑language read and write commands while storing information in a typed, schema‑driven structure. The system extracts objects, relations, and timestamps from free‑form text, deduplicates entries, and maintains explicit links between entities. Every read returns a consistent, queryable state without requiring the model to re‑parse narrative history. Built‑in observability lets developers trace reads and writes, view lineage, and enforce table‑, column‑, and field‑level access controls. The platform can be deployed as a fully managed SaaS service, a zero‑retention SaaS that connects to the customer’s own database, or self‑hosted for full control.
Target Audience
Primary customers are development teams building conversational agents, workflow automation platforms, and enterprise applications that require durable, queryable state across sessions and services.
Features
- Natural‑language write API that parses free‑form text into typed objects and relations defined by a user‑supplied schema
- Structured read API that returns exact facts, supports filters, aggregations, and relational joins
- Automatic deduplication and versioned updates to keep the underlying state current
- Full observability: operation explorer shows reads, writes, trace IDs, and lineage back to source sessions
- Fine‑grained access control at table, column, and field levels
- Export of memory events to external telemetry stacks for monitoring and compliance
- Deployment flexibility: pure SaaS, zero‑retention SaaS with customer‑managed databases, or self‑hosted stack
- Token‑efficient storage that reduces read/write token consumption by over 3× compared to text‑only memory