Skip to content

[Feature]: Evaluate AtomicMemory Core on adebench (open agent-memory benchmark) #73

Description

@adecubed

Area

Core runtime

Use case

I maintain adebench (github.com/adecubed/adebench), an open benchmark for agent memory systems. It already includes ADE Brain, gbrain, Dakera, Memoose and Hindsight. I'd like to add AtomicMemory Core, both as a standalone memory and inside an agent: adebench is adding an agent connector, with Hermes as the first agent, and your native Hermes memory provider fits that directly.

Proposed behavior

An adebench adapter for AtomicMemory Core (self-hosted, via the Python SDK), measuring retrieval and write-path behavior (AUDN decisions, stale or superseded memories).
Two options:
I write the adapter and share the results with you before publishing.
You write the adapter (the interface is documented in the adebench repo), and I run the suite.
Which Core configuration do you consider representative (extraction model, local SLM vs hosted LLM)?

Alternatives considered

Using only your published BEAM and LoCoMo results. adebench adds an independent run on a different corpus and a direct comparison with other memory systems under the same harness.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions