How does optional AI processing work?

OneMira transcribes C1 audio locally. If you configure an AI provider, the desktop app can also run optional title, summary, and Wiki extraction jobs; cloud providers receive the transcript text sent for each selected job.

How do I configure my AI provider?

You can configure multiple AI providers and set one as the default. The desktop application handles all the background routing.

1. Open the OneMira desktop application.
2. Click the Settings (⚙) gear icon in the bottom corner.
3. Select the AI tab in the sidebar menu.
4. Under "New Provider", enter the connection details for your chosen LLM.

Adding a New Provider

To add a new connection, expand the AI Provider section. Fill in the standard OpenAI-compatible fields. If you are using a local provider like Ollama, you can leave the API Key blank. Click Test & Discover to verify the connection before saving.

What does the AI actually do in the background?

After you configure and save a default provider, enable only the workflows you want in the AI Features section.

1. Auto-Title (Best-Effort Background)

When Auto-Title is enabled, the configured provider can replace a timestamp-based session name with a title derived from the transcript. Provider and job status remain visible in the app for review.

2. Wiki Auto-Maintenance & Linting

If you have the Wiki Auto-Maintenance toggle enabled, your chosen LLM acts as a proactive editor for your local knowledge base. You can control how often it runs using the Lint Interval setting. It works in tandem with the Wiki Doctor to perform three critical health checks on your Local Wiki:

• Logical Contradictions (⚠️): The AI analyzes your notes for conflicting statements (e.g., stating a project ships in Q3 in one meeting, but Q4 in another). It ensures your overarching knowledge graph remains consistent.
• Stale / User-Modified Claims (🕒): As conversations evolve, old facts become outdated. The AI flags AI-managed blocks that have fallen out of sync with your most recent transcripts, prompting you to update them.
• Coverage Gaps (✨): The system detects high-frequency entities, acronyms, or project names mentioned across multiple meetings that don't yet have their own dedicated Wiki pages, suggesting you create them for better context linking.

You can always trigger these checks manually by expanding the Wiki Doctor section and clicking the green Scan Now button.

3. Privacy Guardrails

The settings panel identifies when a cloud model receives transcript text. A local OpenAI-compatible endpoint sends requests to the computer or local network URL you configure; its actual data path depends on that endpoint.

Can I manually run prompts against my recordings?

Yes. The mira ai run command lets you execute custom or built-in prompt templates directly against your session transcripts using your configured provider.