User World Model

Concept Overview

Every user has an evolving world model built from the data you register and ingest through the API.
It functions like a compact, continuously refreshed mental model of the user — accessible through deterministic summaries, embeddings, and RAG snippets.

Think of it as user-level memory, structured around 10 key dimensions.

The 10 Dimensions of the User World Model

Below are the key elements that define each user model, along with how they’re captured and applied.

1. Who You Are

Represents the user’s identity and key personal traits.

• Collected through: /register traits (name, timezone, handles)

• Example traits: name, locale, timezone, job title, preferred communication style

• Used for: contextual summaries, timezone-based reasoning, or persona adaptation

2. What You Care About

Captures the user’s recurring interests, goals, and themes of attention.

• Derived from: document titles, event metadata, and content ingestion

• Example: “Ash frequently references product design, scheduling, and deep focus.”

• Used for: proactive alignment and prioritization

3. Who You Interact With

Builds a social and communication graph of relationships.

• Derived from: email headers, chat participants, recurring message senders

• Example: “Ash collaborates most with George (COO), Curren (Chief of Staff), Tom (Engineering Lead).”

• Used for: context handoff, relationship awareness, and team alignment

4. When and How You Act

Models the user’s activity and behavioral rhythm.

• Derived from: timestamps of events, message frequency, and scheduling data

• Example: “Ash is most active from 9 AM–12 PM PT and prefers async messages after 6 PM.”

• Used for: routine prediction, scheduling, and temporal awareness

5. Why You Decide

Learns the underlying motivations, heuristics, and decision patterns.

• Derived from: language cues, decision records, corrections, and task outcomes

• Example: “Ash prioritizes focus and clarity, often optimizing for speed and minimal overhead.”

• Used for: decision prediction, proactive reasoning, and goal alignment

6. Tone, Phrasing, Vocabulary

Encodes the user’s communication fingerprint.

• Derived from: documents, notes, and emails written by the user

• Example: concise, minimal punctuation, preference for imperative phrasing

• Used for: voice mirroring and persona adaptation

7. Circadian Rhythm & Energy Windows

Tracks the user’s natural rhythm and focus cycles.

• Derived from: event timestamps and recurring behavioral clusters

• Example: “Ash’s deep-focus blocks occur 9 AM–12 PM; meetings preferred after 1 PM.”

• Used for: routine prediction, scheduling assistants, wellness agents

8. People Graph & Communication Style

Captures how the user communicates and collaborates.

• Derived from: recurring correspondents and conversational tone analysis

• Example: “Uses short, assertive Slack messages; formal in external emails.”

• Used for: adaptive communication and cross-tool reasoning

9. Document Links & Decision Records

Connects the user’s digital artifacts to their context and intent.

• Derived from: ingested documents, emails, and notes

• Example: “Decision logs and project summaries stored under Notion workspace.”

• Used for: personalized retrieval, reasoning continuity, and multi-tool context

10. Goals, Priorities & Risk Tolerance

Models long-term orientation and preferences in uncertainty.

• Derived from: explicit goals, preferences, and outcomes over time

• Example: “High risk tolerance for technical experiments; prefers predictable team cadence.”

• Used for: proactive alignment and assistant goal inference

Data Flow

Each world model evolves through four key processes:

1. Registration — Define user traits with /register
2. Ingestion — Feed documents, emails, or events with /ingest
3. Embedding — Fastino encodes user context into adaptive latent representations
4. Querying — Retrieve summaries or context via /summary and /query

Example: Building a User World Model

Input Data

{
  "user_id": "usr_42af7c",
  "traits": {
    "name": "Ash Lewis",
    "timezone": "America/Los_Angeles",
    "notes": "Founder/engineer; prefers concise comms"
  },
  "events": [
    {
      "event_id": "evt_1",
      "type": "email",
      "timestamp": "2025-10-10T09:00:00Z",
      "content": "Let’s move stand-up to 2 PM starting next week."
    }
  ]
}

Generated Model Summary

Applications

• Routine Prediction – Agents learn when a user is likely to perform actions.

• Voice Mirroring – Generate text in the user’s tone and phrasing.

• Decision Prediction – Model how a user resolves uncertainty.

• Context Handoff – Transfer context across tools and sessions.

• Personalized Retrieval – Retrieve knowledge aligned to user memory.

Integration Points

To retrieve or update a user’s world model:

Get Summary

Ask a Question

POST /query
{
  "user_id": "usr_42af7c",
  "question": "What is Ash’s typical focus window?"
}

Response

{
  "user_id": "usr_42af7c",
  "answer": "Ash typically focuses from 9 AM to 12 PM PT and schedules meetings after 1 PM."
}

Best Practices

• Keep registration data concise and semantically rich.

• Use consistent user IDs across tools.

• Ingest meaningful events (emails, decisions, notes).

• Periodically refresh context with /summary.

• Use purpose filters to fetch domain-specific views (e.g., work-style, decision-patterns).

Summary

The User World Model transforms fragmented behavioral data into a coherent, queryable profile — enabling agents to reason about users as real, evolving entities rather than anonymous sessions.
It’s the cognitive substrate that powers adaptive, proactive, and context-aware AI behavior.

Next, continue to Ingestion to learn how data flows into the world model.