Heard can wrap command line tools that expose useful terminal output, turn that output into spoken updates, and keep the normal terminal stream visible for direct inspection.
Open source heard run wrapper · Compatibility depends on terminal output · Native adapters remain preferred
(1)The short answer
OpenClaw and Hermes can use the wrapper path when their terminal output is compatible.
With the open source engine installed, launch the normal command through heard run <command> [args...]. Heard starts the child process, preserves its terminal output, and attempts to synthesize useful events for narration from what the process exposes.
This path is intentionally more cautious than native Claude Code or Codex support. Output formats, interactive prompts, hidden state, and alternate-screen interfaces vary, so each tool and version should be verified with a short task before regular use.
01
Keep the command
Use the tool you already run, with Heard added at launch.
02
Preserve output
The terminal stream remains available for direct inspection.
03
Test compatibility
Validate questions, failures, and completion before relying on narration.
(2)Launch
Prefix the existing command with Heard.
If the agent normally starts as agent-name, try heard run agent-name. Add the same arguments you already use. Heard spawns the process and watches the output without replacing the terminal as the source of truth.
(3)Verification
Prove the useful events with one short run.
Ask for a reversible task that produces visible progress, a question, and a clear finish. Confirm the child process behaves normally and compare what Heard says with the terminal output.
Confirm ordinary output still appears in the terminal.
Trigger an interactive question if the tool supports one.
Check that failure and completion summaries remain understandable.
(4)Limits
Use native agent adapters when they exist.
Structured Claude Code and Codex events are more reliable than inferred text. Reserve the universal wrapper for tools without a first-class integration and recheck behavior after major tool updates.
How it works
A careful wrapper setup
01
Wrap
Start the agent with heard run in front of its normal command.
02
Exercise
Run a short task that exposes progress, questions, and completion.
03
Compare
Check every spoken event against the visible terminal result.
OpenClaw and Hermes
Wrapper path
Verify the current CLI output empirically.
Native support
Claude Code + Codex
Use first-class adapters when available.
Terminal hosts
Ghostty + cmux
Hosts do not create agent semantics by themselves.
OpenClaw and Hermes
Questions, answered.
Does Heard have a native OpenClaw or Hermes adapter?
Not a narrating one. The Heard Mac app installs a Hermes plugin that shows Hermes sessions without narrating them. For spoken progress, both tools use the open source engine’s heard run wrapper, so compatibility depends on the output their current versions expose.
Does heard run support every command line agent?
No. It can only interpret output that reaches the terminal in a usable form. Interactive and alternate-screen applications may expose less context.
Will I still see the original terminal output?
Yes. The wrapper tees the child process output so the visible terminal remains available for direct inspection.
