About AtMem
Memory and evidence for agents.
AtMem keeps agent memory under explicit control and records what the agent observed, what context it received, and how a run unfolded. You can inspect those records in the local dashboard.
AtMem works alongside AtBots for optional model-assisted memory proposals and AtFlows for local observability. Explore the three-package portfolio →
Get the source
AtMem is developed in the open. Read the code, report an issue, or contribute on GitHub.
Get started
Install AtMem, connect an agent, and explore its memory and evidence controls.
Product details and availability
Models propose and rank. AtMem alone admits, authorizes, scopes, injects, corrects and deletes.
Agent, workspace, subject, session and run identifiers are security boundaries. Cross-scope reuse fails closed.
Every durable evidence object and its semantic metadata is encrypted at rest. Direct storage inspection yields no session plaintext.
Given only a backup of the evidence store and AtMem, an authorized investigator can reconstruct a run after the agent, its logs and its workspace are gone.
New host integrations begin in shadow mode. Activation is explicit, measurable, and fails closed.
The base installation stays useful without a hosted AI provider. Model selection, remote endpoints and egress are explicit and inspectable.
Forget operations cover and verify canonical, graph, vector and registered derived representations.
Core memory, context, policy and evidence contracts work without OpenClaw imports, identifiers or configuration.
RoadmapA hosted AtMem platform with team workspaces and shared evidence operations. No date. AtMem today installs and runs locally.
Questions or feedback? hello@atmem.ai