FAQ
What is PAM in one sentence?
Section titled “What is PAM in one sentence?”PAM is an open JSON format that lets you move your AI memories (preferences, skills, context) between providers like ChatGPT, Claude, Gemini, and any other AI tool.
Is PAM a storage format?
Section titled “Is PAM a storage format?”No. PAM is an interchange format — it defines how memories are exported and imported between systems. It’s not a database schema or a runtime storage format. Think of it like vCard for contacts or iCalendar for events: a portable representation that any system can read and write.
How is PAM different from just exporting JSON from my provider?
Section titled “How is PAM different from just exporting JSON from my provider?”Provider exports are proprietary and incompatible. ChatGPT’s export has a different structure than Claude’s, which is different from Gemini’s. None of them can be loaded into another provider.
PAM normalizes all of these into one consistent format with typed memories, provenance tracking, integrity hashes, and a defined schema. A PAM file from ChatGPT looks exactly like a PAM file from Claude — that’s what makes migration possible. See Provider Mappings for the specific differences between provider formats.
Does PAM require provider support?
Section titled “Does PAM require provider support?”No. PAM works via official SDK Converters that read provider export files. Providers do not need to natively support PAM. You export your data using the provider’s existing export feature, then an SDK Converter transforms it into PAM format. See the Interop Guide for how converters work.
What happens when a provider changes their export format?
Section titled “What happens when a provider changes their export format?”Converters are versioned. Each converter declares which version of the provider’s export format it supports. When a format changes, a new converter version is created while the old one continues to work for older exports. The spec requires importers to be versioned and resilient to format variations. See Provider Mappings for current converter status.
How do I export my data from ChatGPT, Claude, or Gemini?
Section titled “How do I export my data from ChatGPT, Claude, or Gemini?”Each provider has a different export process:
- ChatGPT: Settings → Data controls → Export data
- Claude: Settings → Account → Export data (provides
chat_messagesas JSON) - Gemini: Via Google Takeout selecting “Gemini Apps”
- Copilot: Via account data export or Microsoft Graph API
Detailed field-by-field mapping for each provider is in the Providers section.
Can I edit my PAM file?
Section titled “Can I edit my PAM file?”Yes. PAM files are plain JSON — you can open, read, edit, and delete anything with a text editor or any JSON tool. Want
to remove a memory the AI inferred incorrectly? Delete the entry. Want to correct something? Edit the content field
and recompute the content_hash. The format is designed to be human-readable and user-controlled.
What memory types does PAM support?
Section titled “What memory types does PAM support?”The built-in types are: fact, preference, skill, context, relationship, goal, instruction, identity,
environment, and project. You can also define custom types by setting type to "custom" and providing a
custom_type field (e.g., "custom_type": "medical_condition"). See the Specification for the full
type definitions.
What about my conversation history?
Section titled “What about my conversation history?”PAM handles conversations separately from memories. Conversations are stored as companion files using a normalized
format that supports features like branching (for providers like OpenAI that allow conversation branches) and multipart
content (text, code, images). They’re referenced from the memory store via conversations_index. See
the Conversations schema for details.
Is PAM open?
Section titled “Is PAM open?”Yes. The specification text is licensed under CC BY 4.0. JSON schemas and reference code are licensed under Apache 2.0. Everything is on GitHub.
Can I use PAM locally?
Section titled “Can I use PAM locally?”Yes. PAM is a JSON format with no server dependency. You can create, validate, and read PAM files entirely offline using standard JSON tooling. See the Quick Start to validate your first file in three steps.
Is PAM production-ready?
Section titled “Is PAM production-ready?”The specification is at v1.0 Published. The JSON schemas validate against Draft 2020-12. Examples and provider mappings are documented. SDK Converters and a reference implementation are under development. The spec is ready for implementation — the tooling is catching up.
Why deterministic hashing?
Section titled “Why deterministic hashing?”Content hashes enable integrity verification without a central authority. Any party can independently verify that memory content has not been tampered with by recomputing the hash using the spec’s normalization rules (trim, lowercase, NFC normalize, collapse whitespace, then SHA-256). This matters for trust — if you receive a PAM file, you can prove its contents haven’t been altered. See Integrity Verification for the full process.
How does PAM relate to MCP?
Section titled “How does PAM relate to MCP?”They are complementary. MCP (Model Context Protocol) is a runtime protocol for AI tool interaction — it defines how tools communicate in real-time. PAM is a data interchange format for persisted user memory — it defines how memories are stored and transferred between systems. They solve different problems and can coexist: an AI tool could use MCP for runtime operations and PAM for importing/exporting user context.
How can I contribute?
Section titled “How can I contribute?”PAM is community-driven. You can contribute by reporting issues, proposing spec changes, building SDK Converters, or improving documentation. See Contributing for details and the GitHub repository for the source.