Long-term memory API for AI apps and agents.
live · privateA multi-tenant REST API and MCP server that turns conversations, documents, URLs and video into structured, queryable memory: facts, entities, relationships and events over time. Apps call save, ask and search, and Hypersave handles extraction, retrieval, versioning and forgetting.
highlights
- 01About 217 endpoints across 36 route modules, a 93-table data model and more than 70 backend function modules: roughly 135k lines of TypeScript over 321 commits.
- 02Hybrid retrieval that fuses facts, chunks, keywords, temporal events and knowledge-graph traversal with reciprocal rank fusion and cross-encoder reranking.
- 03Memory that changes over time: bitemporal facts (when something was true and when it was learned), entity resolution, consolidation, salience and decay, GDPR export and forget.
- 04Redis/BullMQ background workers, distributed rate limiting, audit logs, API-key/JWT tenant isolation and Stripe billing.
- 05CI with type checks, a stricter type pass, about 60 regression test files and a gitleaks secret scan, plus an 11-suite benchmark and offline evaluation harness.
built with
- typescript
- hono
- convex
- redis
- bullmq
- vertex ai
- stripe
- mcp
- aws