Intelligence

The thinking layer

Five interconnected systems that learn, remember, and anticipate. The brain behind every interaction.

Workspace Brain

13-phase synthesis pipeline

Builds comprehensive workspace understanding from every interaction across all channels. Produces 8,800 tokens of pinned context blocks.

$0.25

Full rebuild

$0.10

Incremental

8,800

Token output

  Workspace Brain V2 Pipeline
  ──────────────────────────────────────────────────

  Phase 1    Entity Resolve
  Phase 2-5  Channel Extraction (DB queries)
             Meetings | Email | iMessage | Telegram
  Phase 6-9  Channel Analysis (4x STANDARD AI)
  Phase 10   Merge & Synthesize (1x STANDARD AI)
  Phase 11   Materialize
  Phase 12   6 Pinned Blocks (8,800 tokens)
             Identity | Priorities | Comms | Contacts | Memory | Calendar
  Phase 13   Snapshot + Propagate

  Cost: ~$0.25 full rebuild, ~$0.10 incremental

Memory

Living brain

Importance-weighted recall with semantic search and natural decay. Memories that matter persist; the rest gracefully fade.

Memory formation

Conversations, interactions, and explicit commands create new memories.

Importance scoring

Four-tier classification: low, medium, high, critical.

Semantic embedding

1536-dimensional vectors for similarity search.

Vector storage

HNSW index (cosine, m=16, ef_construction=64).

Retrieval

Semantic search combined with importance weighting and tag filtering.

Lifecycle

Decay for unused memories. Reinforcement for accessed ones.

Episodic memories

Conversation-derived memories with full context preservation and source attribution.

Tag-based filtering

User-defined tags on memories for categorical retrieval alongside semantic search.

Memory decay

Unused memories gradually decay in importance, keeping the system focused.

Reinforcement

Memories that are accessed or referenced get importance boosts.

Skills

Composable user skills

Users define skill profiles that shape AI behavior. Semantic matching finds relevant skills per message, then composes them into the system prompt.

Skill composition

  User Message
       |
  Generate Embedding
       |
  search_user_skills RPC (semantic)
       |
  Top-3 Skill Profiles (ranked by cosine similarity)
       |
  composeSkills()  ->  ComposedSkillProfile
       |
  Inject as <user_skills> XML block into system prompt

Signals

Context signal router

Every extracted signal fans out to living documents, daily context, and entity tracking. Synthesized into a daily digest at 5am.

  context/signal.extracted
       |
       +--- Living Documents (update)
       |
       +--- Daily Context Log (append)
       |
       +--- Entity Interaction Ledger (record)
       |
  5am Cron (MICRO tier synthesis)
       |
  Daily Intelligence Digest (vibe card)

0Brain Phases

0Embedding Dims

0Systems

0Pinned Blocks

Talk to Sentigen

Have questions about what you're reading? The intelligence behind this platform can walk you through it.

PreviousCRM & Entities NextMeetings