Intent Queue for AI Agents

A lightweight queue for mid-task corrections while an agent is running

Problem

When users ask an AI agent to do something, they often send a follow-up message halfway through to correct the original intent. Simple debouncing breaks the 95% of messages that are perfectly fine, while manual interruption breaks flow. Indie developers running multiple agents hit this conflict constantly and have no way to express priority without harming normal input.

Solution

A local or lightweight cloud intent queue that classifies follow-up messages as append, correct, or cancel, then maps them to the agent's interrupt, queue, and cancel signals. Ship a small SDK/middleware that sits in front of OpenAI Assistants, Claude Code, or a custom agent loop, deciding automatically when to interrupt.

Users

Indie developers and small teams using coding agents or automation agents, especially those running several long tasks at once.

MVP

A lightweight middleware that intercepts user input; rules plus a small model classify messages as append/correct/cancel; emits unified interrupt/queue/cancel events; a CLI or browser sidebar shows current queue state; adapters for a few common agent frameworks.

Revenue

Open-source core plus a per-seat subscription for cloud queue and team dashboard; a one-time desktop license for teams wiring multiple agents.

Why now

Agents run longer and mid-course corrections are now the norm, but frameworks only expose low-level interrupt APIs. No one has productized the human side of interruption. An indie dev can build the classification layer with rules and a small model, no training required.