The persistent amnesia of Large Language Models has long been a headache for developers building agentic workflows. Today, a new tool called MemorySync emerged on Hacker News, offering a dedicated API to bridge the gap between ephemeral chat sessions and lasting user knowledge. The project provides a simple endpoint for agents to store and retrieve facts across different conversations.

How the API Works

MemorySync operates on a two-endpoint model: one for storing information and another for searching it. Unlike traditional chat history, which dumps raw conversation logs into a context window, MemorySync extracts specific, lasting facts and stores them server-side. When a new session begins, the agent can query the API to retrieve relevant memories by meaning, effectively allowing the AI to recall user preferences or past interactions without needing the entire previous chat log in its context.

Pricing and Access Model

The service is currently in a free tier phase, requiring no credit card for initial access. Users can generate an API key instantly, which grants 200 writes and 500 searches for seven days. For those wanting to keep their key permanently, adding an email address transitions the account to a free plan with increased limits of 10,000 writes and 1,000 searches per month. The team emphasizes that the playground uses the production API at api.memorysync.io, ensuring that the demo experience mirrors real-world usage.

Key Takeaways

  • MemorySync provides an API for persistent, cross-session memory for AI agents.
  • The system extracts specific facts rather than storing raw chat history.
  • A free tier is available with 10,000 writes and 1,000 searches per month after adding an email.
  • The service uses a production API for its interactive playground.

The Bottom Line

MemorySync tackles the context window bottleneck with a pragmatic, server-side memory layer that deserves attention from agent developers. The free tier’s generous limits and instant access make it a low-risk tool to test persistent memory architectures in production workflows.