AI Briefing
KOSign in

AIPass open-source framework enables AI agents to coordinate via email and dispatch

·2026.10.04 11:10

Key point

The framework uses strict directory isolation and an email/dispatch system to allow agents to report bugs and coordinate tasks without human intervention.

Details

AIPass is an open-source framework designed to solve coordination problems in multi-agent setups by giving agents communication capabilities rather than just improved reasoning. Built over seven months, the system comprises 18 agents and 21,000+ tests, demonstrating that reliability stems from iterative failure and fixing rather than model quality alone.

Communication and Coordination

Each agent operates as a specialist in its own directory with a hard block preventing writes outside its domain. To coordinate, agents use two mechanisms:

  • Email: Drops a message in another agent's mailbox.
  • Dispatch: Sends a message and actively wakes the receiving agent to handle it.

The developer observed that agents spontaneously used this system to file bug reports against each other, such as reporting path resolution failures, which the owning agent would then fix autonomously.

Reliability and Security

Reliability was achieved through extensive testing, with the mail agent alone having 1,460 tests derived from actual failures. Security is enforced by blocking direct file writes to other agents' inboxes, ensuring all communication goes through the mail system. Managers are protected from being woken by workers for grunt work; dispatches to managers are delivered as passive mail.

Implementation

AIPass is CLI-based, built on Claude Code, and runs on Linux, macOS, and Windows. It includes a monitoring agent that logs actions in real time and a watcher that fingerprints errors to dispatch the responsible agent, escalating if issues persist.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.