This TypeScript SDK provides a hosted memory service for AI agents, enabling them to store and retrieve conversational data as structured facts. It targets developers building AI applications who need durable, searchable, long-term memory, offering a significant benefit by allowing agents to recall past interactions and information contextually.

How It Works

The XTrace Memory API ingests raw conversation messages and asynchronously extracts structured facts, artifacts, and episodes. These memories can be tagged and shared across users via groups. The core innovation lies in its encrypted vector search combined with flexible filtering (user_id, group_ids, agent_id, app_id) for precise recall. The recall() function uniquely merges an agent's personal memories with shared group memories in a single query, enabling contextually rich interactions.

Quick Start & Requirements

• Primary install: npm install @xtraceai/memory
• Prerequisites: Node.js 18+ (utilizing native fetch), browser compatibility. Requires API key and Organization ID obtained from app.xtrace.ai.
• Setup: Create a free account at app.xtrace.ai to obtain credentials.
• Documentation: Full details available at docs.xtrace.ai.

Highlighted Details

• Seamless integration with the Vercel AI SDK via @xtraceai/memory/ai-sdk, offering both memory-aware model wrappers for automatic context and ingestion, and memory tools for LLM-controlled access.
• Supports encrypted vector search for secure and private data retrieval.
• Enables cross-user recall through memory tagging and shared group functionality.
• Offers a unified recall() function that combines personal and group memories.

Maintenance & Community

No specific details regarding contributors, sponsorships, or community channels were found in the provided documentation.

Licensing & Compatibility

• License type: MIT.
• Compatibility notes: Works in Node 18+ and in the browser. The MIT license is generally permissive for commercial use and integration with closed-source applications.

Limitations & Caveats

The free tier is subject to rate limits. Synchronous ingestion has a server-side timeout of up to 30 seconds, falling back to asynchronous processing if exceeded. The conv_id parameter is currently a mandatory field for message ingestion.