Post-processing LLMs

Speech models turn audio into text. Post-processing LLMs turn that text into something polished—a clean dictation, a meeting summary, an answer to a question. Hyprcore lets you pick whichever LLM you trust.

Configure in one place

Open Settings → Models → Post-processing. Pick a provider and a model. The same selection drives:

• Dictation post-processing

• Meeting summaries and templates

• Session chat answers

• Action item extraction

You can override per-feature if you want, but most people pick one default and move on.

Provider options

Provider	Models	Where it runs	Cost
Hyprcore Cloud	Claude, GPT, Gemini, Llama (managed)	Hyprcore servers, OpenRouter under the hood	Plan tokens
OpenAI	GPT-4o, GPT-4o-mini	OpenAI API	BYOK
Anthropic	Claude Sonnet, Haiku	Anthropic API	BYOK
Z.AI	GLM models	Z.AI API	BYOK
OpenRouter	Many models (incl. Gemini)	OpenRouter API	BYOK
Groq	Llama Scout, Maverick	Groq API	BYOK
Cerebras	Llama models	Cerebras API	BYOK
Ollama	Whatever you've pulled	Your Mac	Free
LM Studio	Whatever you've loaded	Your Mac	Free
Custom	Any OpenAI-compatible model	Your endpoint	Varies

Picking a model for the job

• Dictation cleanup. Small/fast wins. GPT-4o-mini, Claude Haiku, or Gemini Flash (via Hyprcore Cloud or OpenRouter) all do great. Avoid big slow models—they make every dictation sluggish.

• Meeting summaries. Bigger models produce noticeably better summaries. Claude Sonnet, GPT-4o, or Gemini Pro (via Hyprcore Cloud or OpenRouter). Worth the tokens.

• Session chat. Same as summaries—use a big model.

• Code-heavy meetings. Claude Sonnet handles code in transcripts the most reliably.

Bring your own key

For BYOK providers:

1. Open Settings → Models → Post-processing.

2. Pick the provider.

3. Click Add API key.

4. Paste your key.

Hyprcore stores keys in macOS Keychain—they're encrypted at rest and never leave your Mac unless you make a request to the provider.

Local LLMs (Ollama, LM Studio)

For maximum privacy, run an LLM locally:

Install Ollama or LM Studio

Both are free, separate apps. Install whichever you prefer.

Pull a model

For Ollama: ollama pull llama3.1:8b (or another model). For LM Studio: search and download from the LM Studio app.

Start the server

Ollama runs at http://localhost:11434 by default. LM Studio's local server is at http://localhost:1234.

Point Hyprcore at it

Settings → Models → Post-processing → Provider: Ollama (or LM Studio) and pick the model. Hyprcore tests the connection.

Local LLMs are slower and less capable than the big cloud models. Llama 3.1 8B is a reasonable baseline for cleanup; for summaries, run a 70B if your hardware can handle it.