Vibe Feed
Proactive intelligence
Skill-driven agents produce intelligence cards with a single AI call. 18 card types across 15 specialized agents.
Engine
Card Intelligence Engine
One AI call per card. Skill-driven analysis and formatting in a single pass. 50% cost reduction over the previous dual-call architecture.
1
AI call per card
50%
Cost reduction
18
Card types
Signal Detection (Inngest webhooks, crons, events)
|
Card Intelligence Engine
1. Load skill from DB (card_production_skills, 5-min cache)
2. Load context in parallel (entity profiles, channel history, brain)
3. ONE AI call (skill-specific prompts + Zod schema)
Produces: analysis + headline + subtext
|
Card Orchestrator
skip_formatting: true
(bypasses second AI call)Types
18 card types
| Card Type | Description |
|---|---|
| important_email | Urgent email alerts requiring immediate attention |
| meeting_summary | Post-meeting action items and key decisions |
| email_draft | AI-drafted email responses ready for review |
| email_draft_from_meeting | Follow-up emails drafted from meeting context |
| slack_thread_decision | Decisions detected in Slack threads |
| slack_momentum | Active threads that need your input |
| slack_summary | Channel activity summaries |
| slack_weekly_digest | Weekly summary across Slack channels |
| urgent_slack | Urgent Slack messages needing response |
| urgent_telegram | Urgent Telegram messages needing response |
| telegram_message | Telegram conversation intelligence |
| whatsapp_message | Important WhatsApp conversations |
| imessage_draft_response | Suggested iMessage replies |
| email_automation | Automated email actions based on rules |
| meeting_prep | Pre-meeting participant briefs |
| meeting_debrief | Post-meeting follow-up suggestions |
| task_extraction | Auto-detected tasks from conversations |
| document_suggestion | Relevant documents for current context |
Lifecycle
Card production
1
Signal detection
Webhooks, crons, and events trigger raw data collection.
2
Skill loading
Database-driven skill profile loaded with 5-minute TTL cache.
3
Context assembly
Parallel queries for entity context, channel history, workspace brain.
4
AI analysis
Single call with skill-specific prompts and Zod-validated output.
5
Card creation
Orchestrator creates card with headline, subtext, metadata.
6
Delivery
Vibe feed display. Optional push notification.
Agents
15 agents, 18 skills
| Agent | Card Types | Trigger |
|---|---|---|
| important_email, email_draft, email_automation | Gmail/Outlook intelligence | |
| Meeting | meeting_summary, meeting_prep, meeting_debrief | Meeting pipeline completion |
| Slack | urgent_slack, slack_thread_decision, slack_momentum, slack_summary, slack_weekly_digest | Slack events + crons |
| Telegram | urgent_telegram, telegram_message | Telegram webhooks |
| whatsapp_message | WhatsApp webhooks | |
| iMessage | imessage_draft_response | iMessage intelligence |
| Task | task_extraction | processDetectedTask |
| Document | document_suggestion | Context relevance |
| Email-Meeting | email_draft_from_meeting | Meeting actions |
Architecture
Skill-driven design
Skill definition
Each card type defined by a database row: card_type, domain, prompts, schema, model_tier.
admin.card_production_skills
Adding a new type
Insert one skill row. Gather signal data in agent. Call analyzeAndFormat(). Pass to orchestrator.
1 DB row + 1 agent integration
Talk to Sentigen
Have questions about what you're reading? The intelligence behind this platform can walk you through it.