Which AI models and providers SOPCompass uses
Text features use Anthropic's Claude (model claude-opus-5) through Anthropic's API. The voice interview uses ElevenLabs. Transcription of recordings runs in your browser with Whisper.
On this page
SOPCompass uses three AI components. Each one is named below exactly as it appears in the application code.
Anthropic Claude: every text feature
| Feature | Provider | Model |
|---|---|---|
| Ask | Anthropic API | claude-opus-5 |
| Build with AI (typed interview and drafting the proposal) | Anthropic API | claude-opus-5 |
| Import a document, audio narration | Anthropic API | claude-opus-5 |
| Record a process / Upload a video | Anthropic API | claude-opus-5 |
| Quiz drafting | Anthropic API | claude-opus-5 |
| Capture rewrite | Anthropic API | claude-opus-5 |
| AI sort | Anthropic API | claude-opus-5 |
| Free AI SOP generator (sopcompass.com) | Anthropic API | claude-opus-5 |
Calls go from SOPCompass’s servers to Anthropic’s API. Your browser never talks to Anthropic directly, and the API key never reaches the browser.
Requests are sent with Anthropic’s server side fallback option turned on. If Anthropic’s safety systems decline a request on the primary model, Anthropic can re-run it on its recommended fallback model instead of returning an error. That fallback also runs inside Anthropic’s API, under the same terms.
ElevenLabs: the voice interview
When you choose to talk instead of type in Build with AI, the conversation runs on the ElevenLabs Agents platform. ElevenLabs:
- recognizes your speech,
- produces the interviewer’s spoken replies with the language model configured on the SOPCompass agent in ElevenLabs, and
- reads those replies aloud with ElevenLabs text to speech (model
eleven_flash_v2by default).
The interviewer’s instructions and your workspace context are sent by SOPCompass for each conversation. The words of the conversation are saved back to your interview in SOPCompass. Turning that saved transcript into SOP drafts is done by Claude, the same as a typed interview.
Whisper: transcription in your browser
When you import an audio recording, or a video with narration, speech is turned into text inside your browser using OpenAI’s open source Whisper model (onnx-community/whisper-base.en), run locally with the Transformers.js library. The audio is not sent to a transcription service.
The model files (about 60 MB) download once from the Hugging Face CDN the first time you transcribe, and your browser caches them. That download is a normal file request: none of your audio or content is sent to Hugging Face.
Note: Whisper here is the open source model running on your own computer. It is not OpenAI’s hosted API, and nothing is sent to OpenAI.
Model changes
We may move a feature to a newer model as providers release them. This page is updated when that happens, and the date above shows when it was last checked against the code.
Common questions
Does SOPCompass use OpenAI or ChatGPT?
No. The app's text features call Anthropic's API only. The voice interview runs on ElevenLabs, and transcription of uploaded recordings runs in your browser.
Do you use embeddings or a vector database?
No. Ask finds relevant SOPs with Postgres full text search inside SOPCompass's own database, so no embedding provider receives your content.