What is Mnemosyne?
Mnemosyne maintains a local project memory called The Almanac. It stores distilled knowledge, source references, reliability scores, conflicts, session history and project relationships.
Open source / project memory
Project Mnemosyne
A governed memory and context runtime for AI agents.
Last reviewed 31 August 2026
Active development
Public TypeScript repository; licence not identified in the reviewed metadata.
What problem does it solve?
Agents need useful context across sessions without treating every remembered string as truth or replacing the source files that support it.
How does it work?
The public README describes storage, onboarding, reliability scoring and revalidation, retrieval, decay, structured conflict detection and a transport-neutral tool surface. Untrusted source text stays source evidence rather than runtime instruction.
Key properties
- Project-scoped memory and workspace protection.
- Reliability and revalidation paths.
- Structured conflict detection.
- Separate memory, provenance and action responsibilities.
Mnemosyne does not decide what an agent may do.
Its boundary is memory, provenance and qualified context. Ananke owns action policy and governed execution.
Source and evidence
Inspect the Almanac boundary.
- View source repositoryPublic repository
- Read documentationVerified docs directory
- Inspect the test suiteVerified tests directory
FAQ
Is Mnemosyne just a vector database?
No. Its public project description centres on governed project memory, provenance, reliability, conflicts and transport-neutral context operations.
Can remembered text become an instruction?
The stated boundary is the opposite: untrusted source text remains evidence and is not treated as runtime instruction.