Memory and evidence, under your control

AtMem is a host-neutral Agent Black Box and reversible memory control plane. Keep scoped memories useful across sessions, control which context reaches an agent, and investigate the evidence a supported host captured.

Start with a working example

Install and run your first memory. The Python example runs without an API key or a model download. Then connect an agent and review shadow mode before enabling context injection.

Explore the documentation

• Product portfolio: AtMem, AtBots and AtFlows roles, source repositories and integration boundaries.
• Getting started: installation, first memory and troubleshooting.
• AtBot: optional model-assisted memory proposals governed by AtMem.
• AtFlows: local observability, optional shared login and evidence-review leads.
• Integrations: OpenClaw, frameworks, MCP and delegated providers.
• Reference: commands, HTTP, retrieval, evidence and storage.
• Examples: small runnable workflows and expected results.
• Releases: changes, compatibility and upgrade instructions.

Know what the evidence proves

AtMem retains what supported host boundaries supply. A tool returning successfully is not independent proof that a purchase or message succeeded externally. Missing historical content cannot be reconstructed from a hash. Read evidence and roles before choosing a capture profile.

These are latest-stable guides. For older installations, read that version's release notes and upgrade instructions; full historic guide sets are not published here.