Reference Implementation

The PAM reference implementation is delivered as language-specific SDKs. The Python SDK is the first official release. Each SDK provides the capabilities defined in spec §23:

Available capabilities

1. Platform extractors — Parse exports from ChatGPT, Claude, Gemini, Copilot, and Grok into PAM format. Auto-detect the provider’s export format and map fields to the PAM schema.

2. Converter — Convert provider exports to PAM bundles with pam convert. Outputs valid, schema-compliant memory stores with computed content hashes and companion conversation files. See SDK Converters for details.

3. Validator — Deep validation with pam validate — goes beyond schema checks to verify content hashes, cross-references, temporal ordering, and integrity blocks. See Validation Guide for details.

4. Integrity checker — Verify checksums and consistency rules (content hash normalization, integrity block checksums, ID uniqueness). Runs automatically as part of pam validate --deep (enabled by default).

5. Inspector — Inspect PAM files with pam inspect to get a summary of contents (memory counts by type, relations, conversations index, integrity status).

Planned capabilities

• Signature tools — Sign and verify exports for authentication and tamper detection. Defined in spec §23 but not yet implemented.

Install

pip install portable-ai-memory        # core SDK
pip install 'portable-ai-memory[cli]' # + CLI (pam command)

The Python SDK serves as the canonical example of correct PAM handling.