AI meeting summaries

A 60-minute meeting contains thousands of spoken words, but only a fraction of that dialogue represents lasting intellectual value: concrete decisions made, specific tasks assigned to owners, and core technical trade-offs evaluated.

Expecting team members to read raw transcripts after every call creates cognitive overload. Conversely, relying on generic, open-ended AI prompts often produces verbose, fluffy paragraphs that obscure critical accountability.

Oyma approaches meeting synthesis with deterministic editorial discipline. Proven in src/shared/summary-templates.ts, Oyma provides five battle-tested summary templates engineered to extract high-signal documentation from spoken transcripts, formatted directly into clean Markdown.

Five proven summary archetypes

Rather than forcing you to write complex prompt instructions each time a call concludes, Oyma organizes meeting synthesis into five structured archetypes designed for professional workflows:

1. Executive Summary

Designed for leadership teams and project sponsors who need rapid situational awareness without wading through technical minutiae:

• Core narrative: A concise two-paragraph synthesis explaining the strategic purpose of the discussion and the overarching trajectory.
• Top outcomes: Three bullet points capturing the most significant organizational milestones agreed upon.
• Immediate escalation items: Highlights blockers, budget variances, or cross-functional risks requiring executive intervention.

2. Action Items with Owners and Deadlines

The workhorse template for agile engineering teams, project managers, and product leads:

• Direct task extraction: Every assigned deliverable is converted into a structured Markdown task box (- [ ]).
• Owner attribution: Explicitly attributes accountability to named meeting participants based on verbal commitments.
• Contextual deadlines: Identifies proposed delivery timeframes (such as “by next Wednesday” or “prior to the Q3 code freeze”) without fabricating arbitrary dates.

3. Key Decisions Log

A permanent historical record for architectural reviews, design syncs, and board discussions:

• What was decided: Crisp, affirmative statements documenting finalized agreements.
• Alternatives rejected: Explicitly records options that were evaluated and rejected, ensuring future team members understand why a particular architecture was chosen.
• Underlying rationale: Outlines the technical, financial, or user-experience rationale that drove the consensus.

4. Thematic Breakdown

Ideal for customer discovery interviews, user research sessions, and expansive strategic brainstorms:

• Topical clusters: Automatically groups conversational threads under distinct thematic headings (such as “Usability Feedback”, “Pricing Friction”, and “Enterprise Security Requirements”).
• Direct quote attribution: Selects verbatim participant statements illustrating each theme, paired with clickable audio timestamps.

5. Comprehensive Meeting Record

The all-in-one document template combining high-level narrative, structured decision logs, tabular task matrices, and full attendee rosters. This format is ready to be shared with executive stakeholders or converted using our browser-based Meeting Record Builder.

Iterative summary refinement and regenerating sections

Meeting summaries should adapt as project realities evolve. When reviewing an initial AI summary in Oyma, you are not locked into the first output:

• Regenerate specific sections: If an initial action items list missed a nuance discussed in the final five minutes, you can re-run synthesis specifically for that section without recalculating the entire document.
• Custom instructional modifiers: Append brief stylistic constraints—such as “format as bulleted executive brief” or “focus strictly on infrastructure cost commitments”—to steer the local model toward your exact team expectations.
• Direct manual editing: Because output is pure Markdown inside the live-preview editor, you can tweak action item wording, reorder priorities, or delete tentative musings in seconds.

Local execution with complete privacy

Many AI notetaker apps send your transcripts to multi-tenant cloud clusters to generate summaries, storing your confidential internal strategy on third-party servers.

In Oyma, your summaries can run entirely on your own Mac hardware:

• Powered by local Ollama models: Connect to local AI with Ollama to execute Llama 3, Mistral, or Gemma models directly on Apple Silicon unified memory.
• Zero cloud API costs: Process dozens of hours of recorded meetings each week without paying per-token charges or monthly subscription fees.
• Total air-gap security: Summaries execute without an active internet connection, ensuring complete confidentiality for client briefings, executive compensation discussions, and trade secrets.

If you choose to use cloud frontier models for expansive multi-hour recordings, you can easily bring your own API key for Anthropic or OpenAI, stored securely in the native macOS Keychain.

Seamless integration with your note vault

In Oyma, summaries are not trapped inside a proprietary web dashboard. They land directly in your active Markdown file:

• Formatted with native components: Summaries utilize callouts & highlights, interactive tables, and bidirectional wikilinks.
• Linked to audio timestamps: Action items and quotes link directly back to the exact millisecond in the local recording, allowing you to click and verify context instantly.
• Searchable and portable: The resulting text is indexed by our local full-text search engine and opens natively in Obsidian.

Capture calls with zero-bot meeting recording and transform speech into structured, permanent knowledge with Oyma.