Skip to content
Back to Yap

Explore Yap

Local AI prompts & chains

Clean up, summarize, or transform transcripts with Yap’s built-in text models. Save custom prompts and multi-step chains, keeping each completed result.

Yap’s Prompts screen with the Chains tab selected and a saved Cleanup then Summary workflow using the previous step’s output.

Text models & prompts

Choose a downloaded Gemma model with Yap’s built-in runtime. No separate model app is required. You can also use an external localhost model server if you already have one.

  • Download a text model in Models, choose Use model, and set context length in Prompts → Text model.
  • Choose Cleanup, Summary, or your own prompts from the saved library. Narrow windows use a Saved prompt picker.
  • Save up to 30 prompts with 8,000 characters of instructions each.
  • External servers such as Ollama, LM Studio, and llama-server remain optional. Only localhost endpoints are supported; the external server controls any onward connections.

Process a batch

Run a prompt or chain across selected transcripts or all search matches from History. Each transcript keeps its own saved results.

  • Follow the processing queue in Activity while you keep using Yap.
  • Source transcripts stay unchanged, and every completed chain step is saved.

Generated outputs

Save separate versions of a dictation without replacing its transcript.

  • Each output keeps its input, instructions, and model details.
  • Regenerate using the saved input and prompt with your current model.
  • Read Markdown and copy the source. Outputs remain available with processing disabled.

Previews & refinement

Process an editor draft, review the preview, and choose whether to replace and save it. The original speech transcript is kept.

  • Refine prompt suggests clearer instructions using your selected model.
  • Review suggestions before applying and saving them.

Prompt chains

Build a workflow such as Cleanup → Summary. Each step can use the original transcription, the saved transcript, or the result of a particular earlier step.

  • Save up to 30 chains with 1–10 steps each. New steps use the previous result by default.
  • Reorder steps while preserving their input relationships, or change a step’s prompt and source text.
  • Run a saved chain from a dictation’s Outputs tab. Every completed step keeps its exact input, instructions, model, and place in the chain.
  • Failed or cancelled steps stop the chain while keeping completed results. Editing a chain does not change past outputs.

After dictation

Turn on Run after dictation for a prompt, or Run chain after dictation for a chain. The choice saves automatically. Only one prompt or chain runs automatically at a time.

  • The Automatic badge marks your selection. Dictation profiles can also save a prompt or chain choice.
  • Automatic chains copy or paste only the final result after every step succeeds.
  • If processing fails or is cancelled, retrieve the speech transcript and any completed steps from History or the latest Home result. Nothing is copied or pasted.
  • Incomplete model replies are not saved. Yap can retry with a larger output budget.

Automatic saving

Prompts, chains, text-model settings, Dictionary, and transcript edits save after a short pause. Settings toggles save immediately, and typed settings save when you leave the field or press Enter.

  • Valid changes also save when you leave the page. Invalid drafts pause saving and explain what needs fixing.
  • Failed saves keep your draft and offer Retry or Discard changes. Resolve save errors before quitting or reloading.
  • Use Version history to preview and restore earlier prompts, chains, transcripts, or dictionary.

Performance statistics

Enable optional Performance statistics in Settings → Advanced. Open stats from the latest result, a saved dictation, or a generated output.

  • Inspect transcription time, engine, model, audio length, and transcription speed.
  • For generated outputs, inspect request time, retries, and token counts or generation speed when the local server provides them.
  • Stats stay local with your results. Older results and servers that omit metrics show unavailable values.

Ready when you are

Say your next
sentence.

Free, open-source dictation. Ready when you are.

Download for Windows

Windows 10/11 x64
Internet for initial setup