Heard

About Heard

Heard is a macOS voice layer that connects to agentic coding tools like Claude Code, Codex, and Cursor. It converts their terminal output into spoken summaries, giving you a project-level status instead of scrolling text. The app is open source and free for personal use.

Review

When you run several AI coding agents in parallel, terminal tabs stop scaling. Heard takes the output from those sessions and speaks only what needs your attention-permission prompts, errors, or a decision point. It's built for moments when you're not watching the screen, with a near-silent mode that stays quiet until something goes wrong.

Key Features

• Intelligent speech summaries - Heard produces concise, natural narration from agent output, not a verbatim transcript. It holds session context to decide what's worth saying aloud.
• Hard signal detection - Permission prompts, tool-call failures, and run exits are spoken immediately, regardless of other output. These events cut through routine logs.
• Context-aware filtering - A background pass tracks what's happened so far and learns which patterns matter. It catches decision points in long stretches that look routine but need a response.
• Project-level swarm summaries - When multiple agents run at once, Heard gives each its own voice and then summarizes the group into one clear status update. You hear where the work stands, not five terminals talking over each other.
• Heard Power phone pairing - Pair your phone and the session follows you away from the desk. Audio narrations stream to the phone; you can answer by voice and approve next steps with a tap. The Mac handles all summarization and context, the phone acts as a remote ear and mic.

Pricing and Value

Heard is open source and free for personal use. The phone pairing feature (Heard Power) is described as part of the app, but no separate pricing tier is mentioned. Commercial or team pricing has not yet been defined.

Pros

• Reduces the need to watch multiple terminal windows by surfacing only important events audibly.
• Each agent gets a distinct voice, making it easier to track who said what in a parallel swarm.
• Hard signals for failures and permission prompts prevent missed approval windows that stall workflows.
• Session state stays local on the Mac; no agent data leaves the machine unless you use phone audio relay.
• Open-source codebase allows inspection and community contributions.

Cons

• macOS only - users on Linux or Windows can't run the app.
• Currently works with Claude Code, Codex, and Cursor; agentic tools outside that set aren't supported.
• Summarization may skip low-level log details that some developers want to see in full. Not a fit for those who prefer reading every line of terminal output.

Heard suits developers who routinely run multiple AI coding agents and step away from their desk while work continues. It's also useful for anyone who wants to catch permission prompts or failures without keeping a terminal visible. If your workflow relies on visual scanning of raw logs or you're not on macOS, the tool won't cover those scenarios.

Open 'Heard' Website