Heard Review (2026): Pricing, Features & Honest Verdict

Heard is a macOS voice layer for coding agents. It reads Claude Code, Codex and Cursor output aloud as short spoken summaries instead of terminal walls of text. Useful if you run several agents at once, close to pointless if you run one. Best for: developers juggling parallel AI agents. Price: from $0 (free plan: yes, Apache-2.0 core). Rating: 7.5/10.

What is Heard?

Heard is a macOS menu bar app that sits between your coding agents and your ears. It hooks into Claude Code, Codex and Cursor, watches the event stream those tools emit, and turns the parts that matter into spoken updates. Not a screen reader. Not text to speech bolted onto a terminal. It reads the events, decides what is worth saying, and says that.

The founding team is Kelly, Abdul Rehman and Cruise Chen, and they launched it on Product Hunt in July 2026. The pitch on their own site is blunt about the gap they are filling: “Input tools changed how we talk to agents. Heard handles what they say back.” That framing is the whole product. Dictation tools solved the problem of getting words into an agent. Nobody had solved the problem of a developer running four parallel sessions and having no idea which one just failed.

We found the core is open source under Apache-2.0, and the free tier is not a crippled demo. You can self-host the engine with your own API keys and run a fully local Kokoro voice model, which means zero cost and zero data leaving your Mac. The paid tiers exist because most people do not want to wire up their own TTS provider.

What Are Heard’s Key Features?

Spoken summaries instead of raw transcripts

The thing Heard gets right is compression. Their own example output is “Tests are green. I found one stale import and cleaned it up. The build is ready.” That is one sentence covering what would have been two hundred lines of terminal scroll. Heard listens to the agent’s event stream, works out what changed in context, and speaks only the useful part. If you have ever tried piping agent output into the macOS say command, you know how badly the naive version fails. It reads file paths character by character and never shuts up.

Three listening modes you switch from the menu bar

There is a short mode that gives you hooks and signposts while you work, a fuller briefing mode built for a walk or a drive, and a quiet mode that stays silent until something actually needs a human. That last one is the mode most people will live in. It speaks only for an approval, a blocker, a failure, or a decision waiting on you. Switching between modes is a menu bar click. No prompt tuning, no restart.

Personas and distinct voices per agent

Heard ships several personas with different temperaments: composed and neutral, upbeat, dry and focused, and a richer voice for bigger moments. You can also write your own persona as a Markdown file, which is a nice touch for a tool at this stage. The practical benefit is not personality. It is that when you have three agents running in parallel, giving each one a distinct voice means you know which project is talking before you process a word of it. Heard follows the parallel sessions, voices the highest priority signal, and rolls the rest into compact summaries.

Managed setup or self-hosted, your choice

Heard ships as two things at once. There is the managed macOS app, which handles voices and the summarisation model for you and needs no configuration, and there is the Apache-2.0 engine you can self-host with your own keys and a fully local Kokoro voice. Both are the same product. The self-hosted path means no audio, no code context and no agent output ever leaves your machine, which is the answer to the obvious objection: this tool watches everything your coding agent does. Very few AI-adjacent apps at this price point give you a genuine local option rather than a privacy policy.

Heard Mobile: approve and redirect by voice

The Power tier adds phone pairing. You scan a code once, and live narration follows you to your phone and keeps playing with the screen locked. Hold or tap to talk, and your instruction goes back to the agent still running on your Mac. The speech recognition is code-tuned, which matters more than it sounds: generic dictation mangles technical vocabulary, and an approval that says “merge the auth branch” needs to not come out as “merge the off branch.” Heard also has call-aware auto-pause, scoped pairing and automatic disarm, which suggests they thought about the obvious “my Mac is now voice-controlled by anyone in the room” problem.

How Much Does Heard Cost?

Heard starts at $0. The free tier is genuinely free forever: local Kokoro voices, bring your own AI and voice keys, built-in voices plus custom personas, hotkeys, call-aware auto-pause, and the open source macOS app. Self-hosting is always free. Pro is $15 per month billed monthly, or $12 per month billed annually, and swaps the bring-your-own-keys setup for managed voices and a managed LLM, all-day narration, two premium personas, and cloud sync across Macs. Power is $30 per month monthly or $24 per month annually, and adds Heard Mobile, code-tuned speech recognition, and voice approval or redirection. Managed plans require macOS 13 or later.

Compare that to the input side of the same problem. Wispr Flow has a free Basic tier and charges for Pro. Superwhisper sells a similar dictation product. Those tools cost roughly what Heard Pro costs and do the opposite job, so if you already pay for dictation you are looking at a second subscription in the same budget line. That is the honest sticking point. $15 per month for narration is not obviously cheap when the alternative is $0 with your own keys, and the only thing you are buying is not having to configure a TTS provider.

Who is Heard Best For?

Use Heard if you routinely run three or more coding agents at once and lose track of which one needs you. That is the specific pain this solves, and it solves it well. Also use it if you want to step away from the desk during long agent runs, or if watching terminal output all day is wrecking your focus. The quiet mode plus Power tier voice approval turns a twenty minute agent run into something you can supervise from the kitchen.

Skip Heard if you run one agent at a time. A single session with a desktop notification on completion covers the same ground for free. Skip it if you are not on macOS, because there is no Windows or Linux build. Skip it if you work in an open office, where narration is a non-starter and headphones all day is its own cost. And skip the paid tiers specifically if you are comfortable wiring up your own Anthropic and TTS keys, because the free tier gives you the actual product.

One more group worth naming: teams. Heard is priced and built per developer, with cloud sync across your own Macs on Pro but nothing resembling a team plan. There is no shared configuration, no admin console, no seat management. If you were hoping to standardise agent narration across an engineering team, that product does not exist yet. Buy it as an individual tool or not at all.

Best Heard Alternatives

Pushary

Pushary sends agent permission requests to your phone lock screen so you can approve or deny with one tap. It covers the same “agent is blocked waiting on me” problem from a different angle: notification instead of narration. It works with Claude Code, Codex, Gemini CLI, Cursor and any MCP agent, has a Slack integration and an audit trail, and starts free. If your problem is purely approvals rather than situational awareness, Pushary is the tighter fit and the cheaper one.

Wispr Flow

Wispr Flow is the input half of this equation: voice to text that works in any Mac app, with a free Basic tier and a paid Pro plan. It does not narrate anything back to you. We list it because plenty of developers assume a dictation tool already covers “talking to my agents by voice,” and it does not. If you want both directions you are buying two products.

Claude Code hooks plus the macOS say command

The free DIY route. Claude Code supports hooks, macOS ships say, and forty lines of shell gets you a beep or a short spoken line on task completion. It costs nothing and it works. What you do not get is summarisation, priority ranking across parallel sessions, distinct voices per agent, or anything on your phone. For one agent, this is enough. For four, it is noise.

Final Verdict: Is Heard Worth It?

Heard is a well-built answer to a problem that only some developers have. If you are running one Claude Code session and glancing at it every few minutes, this is a solution in search of your workflow. If you are running four, spread across two projects, and you keep discovering that one of them errored out eleven minutes ago, Heard pays for itself in the first week. The summarisation is the hard part and they got it right, which is not something you can say about most text to speech wrappers.

The pricing is where we would push back. The free tier is strong enough that Pro at $15 per month feels like a convenience fee rather than a product tier, and the genuinely differentiated feature, phone pairing with voice approval, is locked behind the $30 per month Power plan. Voice cloning is still listed as coming soon. Our recommendation: install the free version, run it for a week with your own keys, and only move to Power if you find yourself wanting to approve agents from the other room. Do not start with Pro. Rated 7.5 out of 10, with the caveat that the score assumes you actually run agents in parallel.