About PAM
Mission
Section titled “Mission”Portable AI Memory (PAM) is an open specification for vendor-neutral AI memory interchange. It exists to give users ownership and portability of the context that AI assistants build about them.
Today, every AI provider stores your preferences, skills, and context in a proprietary format that cannot be transferred. PAM defines a common format so your memories can move with you — between ChatGPT, Claude, Gemini, Copilot, local models, and any future provider.
Design goals
Section titled “Design goals”PAM is designed to be:
- JSON-based — standard JSON, readable and writable by any programming language
- Deterministic — content hashing and canonicalization produce consistent, reproducible results
- Verifiable — SHA-256 hashes, integrity blocks, and optional signatures enable tamper detection without a central authority
- Extensible — new memory types, relation types, and metadata fields can be added without breaking existing implementations
- Provider-independent — no dependency on any specific AI provider, platform, or service
What PAM is not
Section titled “What PAM is not”- PAM is not a runtime protocol — it is a data interchange format. For runtime AI tool communication, see MCP
- PAM is not a storage specification — it defines how memories are exchanged, not how they are stored internally. Implementations should use databases and must support export/import using PAM format
- PAM is not affiliated with or endorsed by any AI provider
Governance
Section titled “Governance”PAM is developed as an open-source project. Changes to the specification are proposed via GitHub issues, discussed publicly, and merged via pull request after review. See Contributing for how to participate and the Versioning Policy for how changes are managed.