OpenAI / ChatGPT

The mappings below reflect observed export structures as of February 2026, verified against official OpenAI documentation and community-verified export data. OpenAI does not natively support PAM. These mappings are best-effort compatibility guidance. Provider export formats may change without notice. Importers MUST be versioned.

How to export

1. Go to ChatGPT Settings → Data Controls → Export Data
2. You’ll receive an email with a download link
3. Download and extract the ZIP file

Export files

File	Description
conversations.json	All conversation history as a JSON array with DAG structure
chat.html	Human-readable version (renders JSON via client-side JS)
user.json	Account metadata (id, email, phone, plan)
message_feedback.json	Thumbs-up/down ratings with text descriptions
shared_conversations.json	Conversations shared via public link
tool_messages.json	Tool-related responses and metadata
*.dat	DALL-E image assets (actually PNG files with C2PA metadata)

What you get

Full conversation history with DAG structure. No memories — ChatGPT memories are not included in the export as of February 2026.

Field mappings

Conversation-level mapping

Provider field	PAM field	Transform
id	provider.conversation_id	direct
title	title	direct
create_time	temporal.created_at	datetime.fromtimestamp(v, tz=UTC).isoformat()
update_time	temporal.updated_at	datetime.fromtimestamp(v, tz=UTC).isoformat()
—	provider.name	hardcode "chatgpt"

Message-level mapping

Messages are in mapping[message_id], not a flat array. Traverse by following parent and children references.

Provider field	PAM field	Transform
mapping[k].id	provider_message_id	direct
mapping[k].id	id	generate UUID or use original
mapping[k].parent	parent_id	map provider ID to PAM ID
mapping[k].children	children_ids	map provider IDs to PAM IDs
mapping[k].message.author.role	role	user→user, assistant→assistant, system→system, tool→tool
mapping[k].message.content.content_type	content.type	"text"→"text", "multimodal_text"→"multipart"
mapping[k].message.content.parts[]	content.text or content.parts[]	join parts for text, split for multipart
mapping[k].message.create_time	created_at	Unix epoch float → ISO 8601
mapping[k].message.metadata.model_slug	model	direct if present

DAG structure

The mapping field is a graph, not a list. Some conversations have multiple root nodes or orphaned messages. Walk the graph from root nodes, following children links. Entries where message is null are structural placeholders — skip them.

Timestamps and content

• Some messages have create_time: 0 or null. Use the conversation-level create_time as a fallback.
• parts[] may contain strings, dicts (for images), or null entries. Filter null values before processing.
• DALL-E image data is in message.metadata with a request_id linking to *.dat files in the export.

SDK Converters

PAM import support for ChatGPT is provided by official SDK Converters maintained by the PAM project. See the Importing Guide for general import instructions and the Provider Overview for the full compatibility matrix.