Voice
Audio transcription for voice input.
Send these requests to your own running server. Values stay in this browser.
How this works
Start a server with nikcli serve --hostname 127.0.0.1 --port 4096. Requests go
straight from this page to that address — the browser allows it because loopback counts as a
secure origin, and the server allows it because nikcli-ai.dev is in its CORS
allowlist.
To try things without touching your real sessions or config, start the server with
NIKCLI_TEST_HOME pointed at a throwaway directory — see
the guided tour.
Operations
1 endpoint, generated from the
server's own OpenAPI document. Every path also accepts the
directory and workspace query parameters that select which instance
handles the request.
| post | /voice/transcribe | voice.transcribe |
voice.transcribe | Name | In | Type | Required |
|---|---|---|---|
| directory | query | string | optional |
| workspace | query | string | optional |
{ audio: string format?: string | null }
{ transcript: string error?: string | null }
Example response from schema
{
"transcript": "",
"error": ""
} Generated from the response schema — field values are placeholders, the shape is real.
SDK example
import { createNikcliClient } from "@nikcli-ai/sdk/httpapi"
const client = createNikcliClient()
await client.voice.transcribe({
...
})Try it
Schemas
The 2 schemas reachable from these endpoints. Each is shown one level
deep; nested types link to their own entry below.
Names ending in _2, _3 and so on are the OpenAPI emitter
disambiguating same-named schemas declared by different HttpApi groups — they are distinct
shapes, not duplicates.
{ audio: string format?: string | null }
{ transcript: string error?: string | null }