Skip to main content
Chronologue enables agents and systems to retrieve structured memory context via the MemPort export protocol. The context export interface allows external tools, planner agents, or runtime environments to pull memory traces across time, tempo, and task boundaries. This protocol supports summary-level, full, linked, and delta exports. It serves as the memory transfer layer for hydrating agent runtimes, exporting for review or analysis, and restoring or transferring state between agents or models.

​
Use Cases

  • Hydrate agent runtime with relevant trace context before task execution
  • Share trace summaries with external planning or evaluation tools
  • Reconstruct causal chains from memory (linked_trace_ids)
  • Export historical trace logs for auditing, reflection, or simulation
  • Recover lost agent state by time-bound export and rehydration

​
Context Export Schema

Core fields used in context export queries:
FieldTypeDescription
user_idstringUnique user identifier
start_timeISO 8601Start of time window (inclusive)
end_timeISO 8601End of time window (inclusive)
tempo_tagstringFilter traces tagged with a specific tempo label
task_idstringExport all traces linked to a task
export_formatstringOne of: summary, full, linked, diff
Optional query fields:
  • group_by: day, task, trace_type
  • include_feedback: boolean
  • max_token_count: integer (for summaries)

​
Export Modes

​
summary

  • Token-optimized export for use in LLM prompts
  • Includes trace type, task_id, timestamp, and compressed content

​
full

  • Raw memory trace export (JSON schema)
  • Used for persistence, archival, or external analysis

​
linked

  • Recursively collects traces connected via linked_trace_ids
  • Useful for reconstructing causal chains

​
diff

  • Computes trace differences between two timepoints or exports
  • Used for incremental syncing, audit deltas, or simulation replay

​
Endpoint Specification

​
GET /memport/export

Query Parameters:
ParamTypeRequiredDescription
user_idstringYesUser whose context is being exported
start_timeISO 8601OptionalStart time for filter
end_timeISO 8601OptionalEnd time for filter
tempo_tagstringOptionalFilter by tempo label
task_idstringOptionalFilter by associated task
export_formatstringYessummary, full, linked, or diff
group_bystringOptionalday, task, trace_type
All responses are returned in JSON format. Content-Type: application/json

​
Example Requests & Responses

​
Example: Export Daily Summary

GET /memport/export?user_id=u001&start_time=2025-05-11T00:00:00Z&end_time=2025-05-11T23:59:00Z&export_format=summary Response: 
{
  "user_id": "u001",
  "export_format": "summary",
  "results": [
    {
      "type": "observation",
      "task_id": "goal-2025-05-11-deepwork",
      "timestamp": "2025-05-11T10:00:00Z",
      "summary": "Completed deep work sprint with minor distractions."
    },
    {
      "type": "reflection",
      "task_id": "goal-2025-05-11-deepwork",
      "timestamp": "2025-05-11T20:30:00Z",
      "summary": "Need to block Slack during morning sessions."
    }
  ]
}

​
7. FastAPI Implementation Notes

  • Use @router.get("/memport/export") with dependency injection for auth
  • Validate input with ExportQueryParams Pydantic model
  • Internally call query_traces() and apply formatting logic
  • For summary, include token-count-aware compression if needed
  • Support pagination or limit/offset for long exports
  • Use linked_trace_ids to support linked export mode with trace recursion

​
Redis Acceleration Layer (Optional)

Chronologue supports Redis-backed export acceleration:
Use CaseRedis Key Pattern
Cache recent trace IDsctx:{user_id}:{YYYY-MM-DD}
Cache summary export blocksexport:{user_id}:{scope}:{format}
Ephemeral session memorytrace:{user_id}:{task_id}
When enabled:
  • Exports check Redis cache before querying database
  • Summaries are cached for 15–30 minutes by default
  • Reduces response time for repeated or overlapping queries
Best practices:
  • Use TTLs to prevent memory bloat
  • Store Redis as cache, not source of truth
  • Clear cache on user-triggered edits or trace updates

​
Security and Data Handling

  • All export endpoints require valid API key (Authorization header)
  • User-level access: users can only export their own context
  • Agents acting on behalf of users must be scoped with correct permissions
  • Apply content filtering for sensitive fields (e.g., authored_by, chat_url)
  • Versioned exports can be tagged and checkpointed for reproducibility
⌘I