Agent Inbox

Agent Inbox

The Agent Inbox is a background intelligence layer that monitors your deals, meetings, and pipeline. It surfaces actionable suggestions and waits for your explicit approval before writing anything.

Scheduled agents

• Post-Meeting — After a transcript is saved, proposes Kanban cards and property updates from action items and next steps.
• Deal Monitor — Runs daily. Flags at-risk deals: low health scores, 14+ days without a meeting, close dates within 7 days, or COM completeness high enough to advance the stage.
• Meeting Prep — 60–90 minutes before an upcoming meeting, surfaces prep notes in the inbox.
• Follow-up — After a meeting is analysed, drafts a personalised follow-up email for your approval.
• Daily Digest Email — Morning briefing: today's meetings, overdue items, and closing deals.
• Weekly SE Digest — Every Friday at 4 PM, drafts a weekly update to your SE Manager. Configure the recipient in Settings → Preferences.

Event-triggered agents

• Stage Actions — Fires when you change an opportunity's stage. Creates stage-appropriate task suggestions (e.g. "Define POC success criteria" when moving to POC).

On-demand agents

• Research — On any account or opportunity. Generates 3–5 sales intelligence bullets.
• Next Actions — On opportunities. Recommends 3 next actions based on stage, ARR, close date, and COM completeness.
• Tech Qual — Synthesises the TechQual scorecard by reading all meeting transcripts.
• Stakeholder Map — Identifies stakeholder roles, titles, and relationships from meeting data.

AI analysis: user-triggered

AI analysis (COM synthesis, Three Whys, State Analysis, Presales) does not run automatically after every meeting — this is intentional. Each analysis consumes tokens and time. You trigger it by clicking Run on the opportunity, or by running /analyze-queue in Claude Code.

When you add a transcript or update a deal via MCP tools, a dirty flag is set on that opportunity. This tells Claude Code "something changed — re-analyse when you're ready." The queue drains on your schedule, not automatically.

Claude Code agent automation

When you connect Claude Code via MCP (see Settings → MCP), dirty flags accumulate into an analysis queue. Claude Code checks this queue at session start (if Agent Automation is enabled) and runs pending analyses in sequence. You can also run /analyze-queue manually, or trigger a single analysis with /run-analysis <nodeId> <type>.

All activity is logged to the Agent tab in the notification bell.

The Agent Inbox panel

Click the ✦ sparkle icon in the header. A badge shows pending suggestion count. Each card shows the suggestion type, title, reasoning, and Accept / Dismiss buttons. Nothing is written until you click Accept.

Suggestion types

• Create card — Creates a Kanban card from an action item.
• Update property — Proposes updating a field on an opportunity or account.
• Stage change — Suggests advancing a deal stage.
• Send email — Presents an AI-drafted email with an editable body.
• Health alert — Notifies you of a stale or at-risk deal.
• Meeting prep — Prep notes are ready.
• Research insight — Company intelligence.
• Fill VBC Mantra field — Proposes a value for an empty Mantra field.
• Schedule meeting — Proposes creating a new meeting node.