inioluwa.
🤗 Hugging Face
All projects

Agents · Open-source project

MAOO

Autonomy needs a control loop.

Repository
Eunini/maoo
Stack
Python · Pydantic · SQLite · Docker
Headline
24 evaluation scenarios

A multi-agent orchestrator that perceives, plans, executes and recovers with explicit limits.

Conceptual illustration for MAOO
Conceptual artwork, not a model output or measured figure.

01The problem

An agent that can call tools also needs a way to validate actions, handle failures and know when to stop. MAOO makes those operational decisions inspectable.

02The approach

  1. Separate perception, planning and execution behind typed interfaces, with schema-validated, allowlisted tools.
  2. Track short-term state and SQLite-backed memory. Use bounded retries, argument repair and adaptive replanning when a step fails.
  3. Record structured traces and evaluate behaviour against a mock API with fault injection. An offline deterministic mode supports repeatable development.

03The evidence

  • The repository contains a 24-scenario evaluation harness.
  • Tool contracts, planner policy validation and execution guards have dedicated tests.
  • The optional model integration is separate from the deterministic demonstration mode.

04Scope & status

A portfolio engineering project with a mock execution environment. The 24 scenarios describe the harness, not a claim of universal agent reliability.

Public source is available for independent inspection.

Next projectAgroPredict