Hey Product Hunt-

I’m Matt, maker of OpenDream.

I built OpenDream because even though agent memory has slowly been improving, amnesia across agent sessions continues to be a productivity killer.

Agents don't just "forget" things, they fail predictably.

* agents half-remember old project decisions and fill in the gaps with hallucinations
* outdated context gets leaked into new tasks
* agents fail to see what previous agents already learned and require repeat prompting
* memories become hidden context that is hard to inspect, trust, or correct

OpenDream makes agent memory open, local, portable, and reviewable.

It captures agent activity, turns useful history into saved memory, retrieves only the context that fits the next task, and shows what was selected, skipped, or marked stale.

A few things that make it different:

* Local-first by default
* Open source
* Built for multiple agents and tools
* Source-linked memory
* Reviewable memory changes
* Context retrieval instead of dumping one giant memory file into every prompt

Codex is the most tested with OpenDream. Claude Code, Cursor, Copilot-style repo instructions, Hermes, OpenClaw, and custom runtimes are supported or experimental depending on what each host exposes through hooks, rules, generated context files, or CLI workflows.

I’d especially love feedback from people who use multiple agents on the same project.

What is one thing an agent should have remembered (or remembered incorrectly) that better memory could have helped with?

OpenDream

Open, local-first memory for AI agents (with dreaming)

Open, local-first memory for AI agents (with dreaming)