AI Pulse · Under the Radar

A developer called the project “Mnemos” and dropped it on the MCP subreddit because his agents kept losing track of what they were doing between sessions. He got tired of re-explaining the same context every time he opened a new chat, so he built a memory server and made it open source.

We can’t verify how many people grabbed it, Reddit blocks automated checks now. But the timing matters. Another memory tool, claude-mem, sits at 88,000 GitHub stars and is still climbing by hundreds per day. That tells you builders everywhere are hitting the same wall. Agents are useful until they forget everything you told them yesterday, and nobody wants to babysit a tool that resets its brain every morning.

Why this keeps happening

The MCP standard lets agents talk to external systems, but memory isn’t baked in. So every few weeks someone ships their own fix. Some store notes in a database, others tag conversations, a few try vector embeddings. The approaches vary, but the problem is identical. If an agent can’t remember what you worked on last week, it can’t help you finish the job this week.

For a company running agents in production, this isn’t a nice-to-have. If your support bot forgets a customer’s issue mid-conversation, or your internal assistant can’t recall which supplier you vetoed last month, you’re back to manual lookups. Persistent memory turns an agent from a one-off helper into something that actually learns your business over time. That’s the kind of continuity we wire into the Omni Command Centre, so agents don’t start from zero every session.

The fact that solo builders keep reinventing this wheel means the tooling isn’t solved yet. Until the platforms ship it natively, expect more Mnemos-style projects to land every month.