Open-source TypeScript SDK for AI agents with self-improving memory and production fundamentals built in — temporal knowledge graph, durable runs, cost ceilings, evals in CI. Apache-2.0, runs on Node, Bun, Deno, and the edge.
Eidentic started from a simple frustration: every AI agent I built needed the same two things — memory that actually improves over time, and the production layer (durable runs, cost limits, isolation, evals) — and I kept rebuilding both from scratch. Most tools solve one slice, or hide the production pieces behind an enterprise tier, and almost all of it is Python-first. So I built the thing I wanted: all of it in one composable, fully-open package, native to TypeScript. We publish our benchmarks honestly — including the ones we lose. Apache-2.0, no paid tier.
Our stack
TypeScript end to end, built on the Vercel AI SDK (model-agnostic — any provider through one interface). Ports-and-adapters everywhere: stores (SQLite, libSQL/Turso, Postgres), vector backends (pgvector, Qdrant, LanceDB, Pinecone), and embedders are all swappable. MCP (with OAuth) and A2A for interop; first-class React hooks and a Next.js handler. Runs on Node, Bun, Deno, and the edge. Tooling: pnpm + turbo + tsup + Vitest + Biome, published to npm with OIDC trusted-publishing and SLSA provenance.