Fathom (1)

**Quick judgement** A solid, well-scoped integration for accessing Fathom AI meeting data (calls, transcripts, summaries, and search). It’s most appropriate when users explicitly ask about their own meetings or want to recall/search prior conversations. Treat all outputs as highly sensitive, since they likely contain personal and confidential information. **Key strengths** - Good coverage of Fathom’s core capabilities: listing meetings, fetching transcripts, summaries, and searching by keyword/speaker/date. - Straightforward API-key authentication and clear CLI-style commands that map cleanly to tool calls. - Flexible filters and output formats make it suitable both for direct user answers and for downstream processing. **Main risks & limitations** - **Data sensitivity/privacy:** Call transcripts and summaries can contain PII, corporate secrets, and other sensitive content. Any use in multi-user or shared environments must ensure strict account scoping and avoid exposing one user’s calls to another. - **Misuse via broad queries:** Open-ended prompts like “show me everything from my calls” can surface large volumes of sensitive material; safer to encourage narrowly scoped requests (by date, counterpart, or topic). - **Operational dependencies:** Depends on Fathom’s API availability and rate limits (429 handling). If Fathom is unavailable or throttling, functionality degrades. **Recommended scenarios** - Assist a single authenticated user in reviewing or recalling specific past meetings, with explicit user intent (e.g., “summarize my last Fathom call,” “what action items were in yesterday’s customer call?”). - Search across a user’s historical calls for targeted topics, decisions, or names when the user requests it. - Generate meeting summaries or extract action items from known recording IDs as part of a personal productivity or CRM workflow. **Cautionary notes for deployment** - Only enable for contexts where the API key and Fathom account clearly belong to the active user and are not shared across unrelated users. - Implement guardrails to discourage bulk dumping of all transcripts and to remind users when results may contain sensitive or third-party information. - Ensure logs/telemetry do not inadvertently store full transcripts or sensitive snippets without appropriate protections.