Skip to main content
Follow along with the runnable example. This guide walks through the on-demand agent example in corti-examples — bootstrap your build from it or use it as reference.
Once text is in the editor, you can hand it to a Corti Agentic Framework agent for a transformation the speech-to-text model doesn’t do — for example a minimal copy-edit pass that fixes spelling, grammar, and punctuation without changing wording or meaning. The user triggers it on demand, then the result replaces the editor’s content.
This guide combines the Commands trigger pattern with the Agentic Framework.The prompts used here are mere examples - be sure to test and refine the prompt to support your specific needs.

​
Ensure the agent exists

An agent is created once by name and reused. Look it up in client.agents.list(); create it if missing. See the on-demand agent example in corti-examples for the full ensureAgent implementation. The agent’s behavior lives entirely in its systemPrompt:
Copy-edit system prompt
Cache the returned agent id per API client (e.g. keyed by clientId:tenant) so you don’t list/create on every run. Because behavior is prompt-driven, you can change what the agent does by updating its systemPrompt — no code change.

​
Trigger it on demand

Expose the action as a voice command (and/or a button). Declare a copy_edit command with phrases like "copy edit", "clean up text", or "polish text" in your dictationConfig.commands, then call runAgent(adapter) when that command fires. See the Commands guide for the declare → event → dispatch pattern.

​
Run the pass and write back

Send the editor’s current text as a single isolated message (no contextId — each run stands alone), then replace the editor content via the EditorAdapter with the result. See the on-demand agent example in corti-examples for the full messageSend and write-back implementation.
Guard the write-back: if the agent returns an empty response, leave the user’s content untouched. When a copy-edit agent has nothing to fix, it returns the text unchanged — so writing the result back is safe either way, but an empty result must never clear the editor.

​
Next steps

Agentic Framework

What agents are, how to create them, and the message API.

Second-Pass Agent

Run an agent automatically over a finalized transcript.

Example code

The complete, runnable on-demand agent example in corti-examples.
Please contact us for help or questions.