Ghost in the Machine: DEADLOOP, a Local-First Reliability Monitor for AI Agents

Описание и применение

DEADLOOP – локальный монитор надёжности для ИИ‑агентов, обнаруживает повторяющиеся вызовы инструментов и циклы, выдаёт доказуемые инциденты и позволяет человеку вмешаться без изменения поведения агента.

Задача

AI agents often get trapped in repetitive tool calls, unproductive loops, or circular workflows, wasting tokens and time.

Результат

DEADLOOP detects and reports repetitive agent behavior locally, enabling humans to intervene without external API calls, preserving data privacy and reducing token waste.

Рабочий процесс

• 1. Agent emits tool-call events to DEADLOOP via DeadLoopClient.send_event.
• 2. DEADLOOP ingests events, stores them in SQLite/PostgreSQL.
• 3. Detector logic evaluates traces for loop patterns and session budgets.
• 4. If a finding is detected, DEADLOOP generates an incident with event IDs, metrics, confidence, and uncertainty.
• 5. Human reviews incident, can acknowledge, dismiss, request pause, or approve pause.
• 6. If pause approved, agent execution is halted without claiming control.

Доказательства

DEADLOOP ingests its tool-call trace and answers: Is this repetition, or legitimate retrying?

Источник

The core is fully deterministic; an optional open-weight model only explains the findings. It never changes a finding or executes an intervention.

Источник

It runs offline. Detection, risk scoring, ingestion, and incident management are deterministic and local. No API key is required for the core product to function.

Источник

The sample simulator generates real execution events through the same ingestion, persistence, detection, and incident paths as production.

Источник