AgentMemory – Governed shared memory for agent fleets

Shared memory with built-in judging to prevent context rot across multiple agents

Problem

Developers manually handoff context between agents (e.g., Codex and Claude), leading to memory pileup and quality decay. Existing cross-platform memory solutions lack governance and require manual cleanup.

Solution

A shared memory service with per-group scopes and a judge agent that periodically evaluates memory quality, automatically discarding stale or low-quality entries. API or MCP access.

Users

Developers and AI teams running complex projects with multiple agents.

MVP

Basic memory store and retrieval API; scope isolation; simple judge rules (time decay, duplicate detection); CLI to inspect and clean memory.

Revenue

Open core + paid cloud (hosting, advanced judge models, team features), usage-based.

Why now

Multi-agent workflows grow but memory management is chaotic; indie devs can offer lightweight governance while big vendors build bloated suites.