FishMem

Memory infrastructure should stay understandable.

FishMem is an open memory layer for AI applications. We build the engine, SDKs, Desktop app, hosted service, and documentation as one portable contract.

The goal is not to remember everything. It is to preserve the right evidence, form useful durable knowledge, and make every retrieval explainable enough to improve.

Why FishMem exists

Long-running agents need more than a transcript window. They need scoped records, correction history, source-preserving documents, and retrieval that can be inspected when an answer is wrong.

FishMem keeps raw sources authoritative and treats summaries, profiles, chunks, embeddings, and graph projections as rebuildable views. That boundary makes migration and debugging possible.

  • Durable memory records with provenance and history
  • Source-preserving document retrieval
  • Cloud, self-hosted, and local Desktop surfaces
  • Versioned APIs and portable namespace exports

How we build

We publish the open engine under Apache-2.0 and use the same core contract across deployment surfaces. Hosted-only concerns—organizations, billing, managed workers, and infrastructure—remain explicit rather than leaking into memory semantics.

Benchmarks are published with harness details and limitations. A result is evidence for one workload, never a promise that every application will see the same outcome.

Read the benchmark evidence

Operator

FishMem is operated, contracted, and paid through BLENDDUCK LLC. Product support and commercial questions are handled through FishMem channels.

Legal notices, contracts, invoices, and payment records identify BLENDDUCK LLC as the operating entity behind FishMem.

Contact FishMem