Shared memory for you and your AI

Every conversation starts from scratch. When platforms add memory, it's a black box you can't see or control. Atomic is different — a shared knowledge base where both you and your agents read and write. You stay in control.

A feedback loop, not a black box

Both you and your AI agents contribute to the same knowledge base. What one writes, the other can find.

  1. Remember

    Your agents remember

    They store what they learn via MCP. Preferences, decisions, project context — anything worth retaining across conversations. Knowledge accumulates automatically.

  2. Curate

    You curate

    See everything your agents stored. Edit, organize, add your own knowledge. Shape what they know. You're not monitoring a black box — you're actively shaping your agent's context.

  3. Shared

    One knowledge base, two contributors

    Agents search before answering. You browse and enrich. Both sides draw from the same source — what one writes, the other can find.

Four tools. That's the entire API.

Atomic exposes a minimal MCP surface. No bloat, no complexity. Agents get exactly what they need.

semantic_search

Search memory for relevant knowledge before responding

read_atom

Read the full content of a specific memory

create_atom

Remember something new worth retaining

update_atom

Revise existing knowledge when it's gone stale

Works with every MCP client

Standard Model Context Protocol over Streamable HTTP. Connect any agent that speaks MCP.

  • Claude Code
  • Claude Desktop
  • Cursor
  • Windsurf
  • Any MCP Client

Your memory, your rules

See everything

Browse what your agents remember. Canvas view shows connections. Search finds anything.

Edit and enrich

Add notes, articles, and context your agents can discover. Curate what they've stored. Delete what doesn't belong.

Self-hosted

Runs on your machine or your server. Open source. Your data never leaves your infrastructure.

Give your AI agents memory

Open source. Self-hosted. You stay in control.