Impri

Impri provides a critical human-in-the-loop capability for autonomous agents, ensuring no action is taken without explicit human consent. Key features include: • Agent action proposals requiring human review • Approval, rejection, or editing from a unified inbox (web, mobile, Slack, Discord, Telegram) • Data-dependent gate for execution, preventing skips • Open-source (MIT), self-hostable core • Supports MCP (Message-Channel Protocol) and REST APIs This platform acts as a safety layer, not an agent itself, by queueing proposed actions. Your agent senses the world through 'watchers' that poll data sources like RSS or Reddit, then proposes an action (e.g., sending an email, posting a comment, running a command) with a preview. The human user then makes a decision, which is returned to the agent via webhook or polling, only allowing execution if approved. Impri is designed for fail-closed operation, meaning if there's any doubt, no action is taken. Decisions are immutable and audited for accountability. It offers an agent-first API for seamless integration and allows full core self-hosting with `docker compose up`, ensuring data ownership and exportability. Ideal for engineering teams, developers, and product managers who deploy autonomous agents and require rigorous oversight and control over their outputs, ensuring that all agent activities align with human intent and policy.