Hey Product Hunt 👋 I built earshot because I kept seeing the same failure mode in coding agents: they are often very capable, but not always very attentive. They guess when a request is ambiguous. They modify files outside the intended scope. They forget preferences. They report success without making it clear what was actually verified. earshot treats those behaviours as product bugs. It is an open-source terminal coding agent with: - Support for 21 providers and 1,000+ models - A full interactive terminal UI - Permission-aware tools - Scope contracts - Preference memory - Sessions, rewind, and undo - MCP, skills, hooks, slash commands, and subagents - Headless and ACP modes - End-of-turn verification The most important feature is not a model integration. It is the interaction model: earshot should ask when it needs clarification, explain what it is about to do, and be honest about the result. It is still pre-1.0, so this launch is primarily about finding people who want to try it with real providers, real repositories, and real terminal workflows. Try it here: npm install --global @raegent/earshot earshot doctor I’d especially love feedback on: 1. Which provider and model you use 2. How the permission flow feels 3. Whether the agent asks the right questions 4. What it gets wrong in your terminal workflow Thanks, and please let the thoughts come in! :)

earshot

A terminal coding agent that actually listens.

A terminal coding agent that actually listens.