AI product
Harness Memory is a long-term memory system built into HarnessRouter for AI agents. It loads memory when a session starts, lets an agent search and add records through tools during a turn, and records what mattered when the turn ends so the agent can recall it in later sessions. Agents running on different harnesses, such as Claude Code and Codex, can share one memory.
Each person using an agent can receive a separate memory whose records are kept apart and can be erased independently. Recalled memory is passed as data, fenced off from the agent's instructions and labeled with who wrote it; agents are not given an erase tool. People and agents receive access per memory, and the Memory page provides Data, Schema, and Settings views for inspecting, editing, deleting, and configuring memories.
The system supports memory engines based on short facts, temporal graphs, consolidated facts, or Markdown files: Mem0, Graphiti, Hindsight, and ReMe. It defines the Harness Memory Protocol as part of the Unified Harness Protocol, allowing memory services to plug in as providers. The page identifies the listed engines as Apache-2.0 or MIT licensed and offers self-hosted deployment.
1 use taken from transcripts — each links to the moment in the video.
Long-term memory built into Harness Router for AI agents, where each agent gets its own memory automatically and agents on different harnesses like Claude Code and Codex can share one memory, with recall treated as data only people can erase.
1 in the library.