Remember

API Reference

Store Memory

POST /v1/users/{userId}/memories
Content-Type: application/json

{
  "content": "string",
  "type": "fact" | "note" | "message",
  "authorType": "agent" | "user"
}

Memory Types

Response

{
  "success": true,
  "memoryId": "uuid"
}

Duplicates (same user + content) are skipped automatically.

Retrieve Memories

GET /v1/users/{userId}/memories?query={text}&limit={n}

• With query: semantic search (vector similarity, falls back to keyword)
• Without query: returns recent memories

Response

{
  "success": true,
  "memories": [
    {
      "id": "uuid",
      "content": "string",
      "type": "fact",
      "authorType": "agent",
      "createdAt": "2024-01-01T00:00:00Z",
      "similarity": 0.85
    }
  ],
  "searchMethod": "vector" | "keyword" | "recent"
}

Context Window Preservation

When the context window is reaching its limit, store a comprehensive session summary. This is critical for maintaining continuity.

What to Capture

Create a detailed record containing:

Session Overview

• What task or problem was being worked on
• The user's original request and goals
• Current progress and completion status

Technical Context

• Project structure and relevant directories
• Files read, created, or modified (with paths)
• Key code patterns, architectures, or conventions discovered
• Dependencies, configurations, and environment details
• Database schemas, API endpoints, or data structures involved

Decisions and Rationale

• Design decisions made and why
• Trade-offs considered
• Alternatives rejected and reasons
• User preferences expressed during the session

Current State

• What was just completed
• What is currently in progress
• Blockers or issues encountered
• Error messages or debugging findings

Next Steps

• Pending tasks not yet started
• Follow-up items identified
• Questions that need user input
• Planned approach for remaining work

Storage Format

Store as a note with authorType agent. Structure the content clearly:

{
  "content": "## Session Context\n\n### Task\n[Detailed description of what we're working on]\n\n### Progress\n- Completed: [list with details]\n- In progress: [current work]\n- Pending: [remaining items]\n\n### Technical Details\n- Project: [path and structure]\n- Files modified: [list with purposes]\n- Key patterns: [architectural notes]\n\n### Decisions\n- [Decision]: [Rationale]\n\n### Blockers\n- [Issue]: [Status and findings]\n\n### Next Steps\n1. [Specific action]\n2. [Specific action]",
  "type": "note",
  "authorType": "agent"
}

When to Trigger

Store context preservation when:

• Context usage reaches 75% of the window
• Before starting a task that may exceed remaining context
• When switching between major subtasks
• Before any operation that might truncate history

Usage Guidelines

1. Store facts when user shares preferences or personal info
2. Store notes to summarize important context
3. Store comprehensive context when nearing context limit
4. Retrieve memories at conversation start for continuity
5. Search relevant memories when context would help