OpenDream, an open-source memory system for AI agents, lets multiple agents share and review context across sessions while keeping data local and auditable.

What happened: OpenDream is an alpha-stage tool that stores agent activity, memories, and context in a local workspace that multiple agents (Codex, Claude Code, Cursor, and others) can access and contribute to across sessions. The system includes a 'dreaming' feature—background memory review and cleanup—and an observability UI that shows recent activity, audit records, and review decisions without requiring users to read raw JSON.

Why it matters: Agents currently lose useful context between sessions or rely on closed, opaque memory systems. OpenDream makes agent memory open, portable, and reviewable, so teams can see what an agent learned and reuse it across tools and projects. Reviewability is described as part of the product goal because memory can harm performance when stale or incorrectly scoped.

What to watch: OpenDream is currently alpha software; Codex is the most tested integration path, while other agents depend on each host's hooks, rules, or CLI support. The project includes CLI and UI demo galleries, compatibility matrices for Claude Code, OpenAI Codex, Cursor, GitHub Copilot, and Hermes Agent, and installation via `uv tool install opendream` or from Git.