# Heard vs. Coding Agent Notifications Source: https://heard.dev/compare/coding-agent-notifications Compare Heard’s session-aware spoken agent updates with terminal bells, desktop notifications, and generic text-to-speech. Comparison · Agent notifications # Heard is an attention layer, not a louder terminal bell. Traditional notifications tell you that something happened. Heard is designed to say which agent changed state, what matters, and whether a response is required. Transparent comparison · No universal “best” claim · Updated October 2026 Updated October 6, 2026 The short answer ## Use standard notifications for simple state changes; use Heard when context and parallel sessions matter. A terminal bell or desktop banner is cheap, familiar, and often enough for one process with a binary done-or-not-done state. It is also easy to miss and usually requires reopening the source window to understand what happened. Heard adds natural-language compression, session identity, a filter for what is worth saying, and voice replies on Power. That is more useful for long or parallel agent runs, but it adds another application and, for managed features, an optional hosted processing path. ### Terminal bell Lowest complexity; minimal context. ### Desktop banner Scannable and persistent; still screen-dependent. ### Generic TTS Audible but often too literal and noisy. ### Heard Session-aware compression with voice reply on Power. (1) Best for simplicity ## Terminal bells and native banners. Choose a bell or desktop banner when the only question is whether one process finished. These tools are easy to trust because they do little, and they preserve the original window as the place to investigate. - Almost no setup. - No semantic summary. - Can be missed among other notifications. - Usually requires returning to the screen. (2) Best for literal access ## Generic text-to-speech. Generic TTS can make terminal output audible but normally lacks agent semantics and prioritization. It may read repetitive tool churn, code fragments, or logs that are difficult to understand in audio. - Works with any available text. - Can preserve exact wording. - High risk of verbosity and alert fatigue. - No built-in session salience. (3) Best for agent supervision ## Heard’s session-aware voice layer. Heard judges which events deserve speech, maintains recent context, distinguishes active sessions, and speaks questions, approvals, results and failures rather than every tool step. Native Claude Code and Codex integrations provide richer signals than the universal wrapper. - Questions, blockers, failures, and results are prioritized. - Parallel sessions retain identity. - Managed and open source processing choices. - Requires macOS 14 or later today. (4) Decision rule ## Pay for context only when context saves attention. If a simple bell reliably brings you back at the right moment, keep it. If you repeatedly reopen several sessions just to discover that nothing needs you, or miss the one that does, Heard addresses a materially different problem. At a glance ## Four ways to hear from an agent. A summary of the notes above. Exact behavior depends on your terminal, notification settings, and integration path. - **Semantic summary:** Terminal bell: No; Desktop banner: No; Generic TTS: No; Heard: Yes - **Parallel session identity:** Terminal bell: No; Desktop banner: No; Generic TTS: No; Heard: Yes, by integration - **Zero new app:** Terminal bell: Yes; Desktop banner: Usually; Generic TTS: Varies; Heard: No - **Persistent visual history:** Terminal bell: No; Desktop banner: Yes; Generic TTS: No; Heard: No - **Voice response:** Terminal bell: No; Desktop banner: No; Generic TTS: No; Heard: Heard Power Product facts ## Feature comparison Semantic summary **Heard** Generic tools usually read or signal rather than interpret. Parallel session identity **Heard** Depends on the integration path. Zero new app **Terminal bell** Use the tools already in your terminal. Persistent visual history **Desktop banner** Audio is transient by nature. Voice response **Heard Power** Press right ⌘ and Parrot sends your reply to the session you name. Questions ## Frequently asked questions ### Is Heard better than every coding-agent notification tool? No. A terminal bell or banner is simpler and may be sufficient for a single predictable process. Heard is designed for context-rich and parallel agent supervision. ### Why not send every terminal line to text-to-speech? Literal output is often repetitive, code-heavy, and difficult to follow by ear. Heard aims to compress the stream into material state changes. ### Does Heard keep a persistent notification history? Heard’s core value is spoken awareness and session context. Use the agent, terminal, source control, and test output as the authoritative record. Keep going ## Related Heard resources [**terminal-notifier for coding agents** Set it up for Claude Code and Codex, and see where it stops.](https://heard.dev/compare/terminal-notifier-for-coding-agents) [**Text-to-speech for Claude Code** VoiceMode, claude-code-narrator, a say hook, and Heard.](https://heard.dev/compare/claude-code-text-to-speech) [**Notification design guide** Build a five-level agent alert hierarchy.](https://heard.dev/guides/coding-agent-notifications) [**Compatibility** Review native, observed, wrapper, and terminal-host support.](https://heard.dev/compatibility) [**Open source engine** Inspect how Heard processes agent events.](https://github.com/heardlabs/heard) Put it into practice ## Give your agents a voice. [Get started with Heard](https://docs.heard.dev/quickstart)