Ask your meeting or note

When reviewing a 90-minute technical debrief, a complex client discovery call, or an extensive research document, you rarely need to re-read the entire file from beginning to end. Instead, you have specific, targeted questions:

• “What was the agreed deadline for the database schema migration?”
• “Did the client express reservations regarding our proposed SLA tiers?”
• “Summarize the three security objections raised during the third quarter review.”

Many AI-powered note apps attempt to address this by indexing your entire personal workspace into a single massive vector database. But global search chatbots frequently hallucinate, pull outdated snippets from deprecated projects, and blur confidential boundaries across unrelated client files.

Oyma takes a grounded, high-precision architectural approach. Implemented in src/renderer/workspace/AskWorkbench.ts and ask-context.ts, the Ask feature gives you a conversational inquiry interface strictly bound to your active note, transcript, or highlighted selection.

High-precision scoped context: Zero cross-project leakage

The defining virtue of Oyma‘s Ask workbench is strict context scoping:

1. Explicit document boundary: When you open the Ask panel, the context window is constructed exclusively from the document currently in focus. It does not scan other folders, past archives, or unrelated client records.
2. Precision text selection targeting: If you only want to interrogate a specific section of a transcript (such as a 15-minute debate on pricing tiers), highlight that text. The Ask engine automatically narrows its operational prompt to that exact excerpt.
3. Deterministic grounding: The model is instructed to answer strictly based on the provided text. If a detail was not discussed during the meeting or recorded in the note, the model clearly states that the information is absent rather than guessing or fabricating answers.

This disciplined boundary eliminates hallucinations and ensures that confidential notes from Project A never inadvertently leak into queries regarding Project B.

Fluid conversational investigation

Interacting with your documents feels as natural as consulting a knowledgeable research colleague:

• Follow-up clarifying inquiries: Ask follow-up questions in sequence: “Who owned that deliverable?” followed by “Did they state any dependencies?” The active conversation maintains localized conversational memory.
• Timestamped source citations: When answering questions against an audio transcript captured via zero-bot meeting recording and live transcription, every claim includes a clickable timestamp (such as [32:15]). Clicking the citation jumps the audio player straight to that moment in the call.
• One-click note insertion: When you uncover a valuable insight or formulate a polished summary, click Insert into note. Oyma formats the answer as clean Markdown—complete with callouts and tables—and appends it directly into your active document.

Run locally with Ollama or connect frontier models

Just like our AI meeting summaries, the Ask workbench is built on open, sovereign technology:

• 100% on-device with Ollama: Connect to local AI with Ollama to run Llama 3, Mistral, or Gemma directly on your Mac. You can interrogate sensitive transcripts on airplanes, in confidential client clean-rooms, or behind strict enterprise firewalls with zero internet connection.
• Bring your own API key: If you are analyzing a multi-hour workshop with an extensive transcript, you can seamlessly bring your own API key for Anthropic Claude 3.5 Sonnet or OpenAI GPT-4o. Keys are encrypted securely in the native macOS Keychain.

Quick prompt chips for common interrogation tasks

To accelerate your analysis, the Ask workbench includes pre-configured prompt chips above the query input:

• “List all action items with owners:” Quickly extracts commitments without drafting custom instructions.
• “What concerns or objections were raised?”: Isolates points of disagreement or unresolved technical friction.
• “Summarize decisions agreed upon in bullet points”: Produces a clean decision record ready for sharing.
• “Extract key deadlines and timeline milestones”: Synthesizes dates, code freeze targets, and client deliverables into chronological order.

Clicking any prompt chip instantly submits the query against the active document context, delivering formatted markdown answers in seconds.

Enhancing daily research and meeting velocity

The Ask workbench bridges the gap between passive reading and active knowledge synthesis:

• Meeting prep: Open last week’s meeting record and ask: “What action items did I commit to delivering before today’s sync?”
• Literature review: Ingest an academic paper or technical whitepaper transcribed from a lecture via audio file transcription, and query the core methodology in seconds.
• Decision verification: Audit conflicting recollections across team members by asking the transcript for exact statements regarding architectural scope.
• Confidence through audio cross-checking: Because answers include audio timestamp links, you can verify ambiguous phrasing against the spoken recording with a single click.

Combine grounded conversational analysis with our live-preview Markdown editor to transform static notes into interactive intellectual partners in Oyma.