Cohere STT for OpenClaw

Unofficial, community-maintained OpenClaw plugin that adds the Cohere audio transcription provider.

• Package name: openclaw-cohere-stt
• Extension/plugin id: cohere-stt
• Media-understanding provider id: cohere
• License: MIT
• Required auth: COHERE_API_KEY or an auth profile referenced from tools.media.audio.models[].profile

The extension registers the cohere media-understanding provider, so OpenClaw config uses tools.media.audio.models and the normal provider auth flow.

Requirements

• OpenClaw >=2026.4.11
• Cohere API key via COHERE_API_KEY or an auth profile for provider cohere
• Audio files supported by the Cohere transcription endpoint: flac, mp3, mpeg, mpga, ogg, wav

Install

Once published, install it like any other external OpenClaw plugin:

openclaw plugins install openclaw-cohere-stt

Recommended auth setup

Use COHERE_API_KEY in the runtime environment. This is the simplest path and matches the plugin manifest's auth metadata.

export COHERE_API_KEY="your-cohere-api-key"

Minimal audio config

{
  tools: {
    media: {
      audio: {
        enabled: true,
        models: [{ provider: "cohere", model: "cohere-transcribe-03-2026", language: "ja" }],
      },
    },
  },
}

Example with supported settings

{
  tools: {
    media: {
      audio: {
        enabled: true,
        providerOptions: {
          cohere: {
            temperature: 0.2,
          },
        },
        models: [
          {
            provider: "cohere",
            model: "cohere-transcribe-03-2026",
            language: "en",
            baseUrl: "https://api.cohere.com/v2",
          },
        ],
      },
    },
  },
}

Alternative auth setup: auth profile

If you want to avoid a provider-wide env var, create an auth profile and point the audio model entry at it.

tools.media.audio.models[].profile:

{
  tools: {
    media: {
      audio: {
        enabled: true,
        models: [
          {
            provider: "cohere",
            model: "cohere-transcribe-03-2026",
            language: "ja",
            profile: "cohere:default",
          },
        ],
      },
    },
  },
}

auth-profiles.json:

{
  "version": 1,
  "profiles": {
    "cohere:default": {
      "type": "api_key",
      "provider": "cohere",
      "keyRef": {
        "source": "env",
        "provider": "default",
        "id": "COHERE_API_KEY"
      }
    }
  }
}

OpenClaw also supports inline key instead of keyRef, but env-backed refs are easier to rotate and keep out of committed config.

What the plugin sends

The current Cohere transcription docs confirm these request fields:

• file
• model
• language
• temperature

This plugin intentionally keeps the request surface to that documented set. It does not add prompt injection or undocumented multipart fields.

Config reference

OpenClaw behavior notes

• This is not realtime STT. It handles pre-recorded audio attachments through tools.media.audio.
• Transcript output is injected back into the normal OpenClaw reply pipeline as {{Transcript}} and an [Audio] block, same as other audio providers.
• Provider auth follows the normal provider resolution order. In practice, COHERE_API_KEY or tools.media.audio.models[].profile are the useful options for this plugin.
• tools.media.audio.providerOptions.cohere.temperature is merged from the global audio config and the individual model entry, with the entry taking precedence.

Validation

pnpm install
pnpm run check

References

• Cohere API overview: https://docs.cohere.com/reference/about
• Cohere API reference: https://docs.cohere.com/reference/create-audio-transcription

Support

If this plugin is useful, you can support maintenance here:

• https://github.com/sponsors/livingghost