How Identity OS Works¶

Architecture¶

Your agent sends observations. The engine processes them through a deterministic pipeline. The output is an ExecutionContract.

graph LR
    A[Your Agent] -->|observation| B[Identity OS Engine]
    B --> C[ExecutionContract]
    C -->|controls| A

    subgraph Engine Pipeline
        B --> M[Mode EMA]
        M --> E[Energy]
        E --> S[Stress]
        S --> AR[Arbitration]
        AR --> D[Drift]
        D --> ST[Stability]
        ST --> MEM[Memory + Self-Model]
    end

    style B fill:#6c63ff,color:#fff
    style C fill:#3fb950,color:#fff

Key principle: The engine is deterministic. No LLM inference. Same input + same state = same output.

The Pipeline¶

Step	Module	What It Does
1	Mode EMA	Updates behavioral mode strengths from observation
2	Energy	Calculates energy cost and recovery
3	Stress	Evaluates stress level (LOW → MED → HIGH → OVER)
4	Arbitration	Weights modes by stress level
5	Composites	Detects active mode pairs
6	Drift	Compares current behavior to baseline anchor
7	Stability	Computes behavioral consistency index
8	Memory	Updates episodic memory + self-model + narrative
9	Contract	Builds ExecutionContract with 5 control dimensions

5 Control Dimensions¶

graph TD
    EC[ExecutionContract]
    EC --> A1[Action Space]
    EC --> A2[Timing Control]
    EC --> A3[Goal Filtering]
    EC --> A4[Trust Profile]
    EC --> A5[Personality]

    style EC fill:#6c63ff,color:#fff

Learn More¶

The Four Pillars¶

1. Behavioral Modes¶

Your agent's behavior emerges from 7 coexisting dimensions (Perception, Exploration, Order, Assertion, Connection, Identity, Stress Response). Each mode activates based on the observations you feed in.

You don't need to configure these manually — just send observations, and the engine determines the optimal mode balance.

Learn more →

2. Stress-Adaptive Constraints¶

When your agent encounters failures, retries, or contradictions, stress rises automatically through 4 levels: LOW → MED → HIGH → OVER. As stress rises, the system tightens behavioral constraints — fewer allowed actions, more conservative decision style.

Learn more →

3. Drift Detection¶

Every behavioral change is classified on a 4-level scale: D0 (noise) → D1 (context) → D2 (adaptive) → D3 (corrupt). Critical drift triggers automatic rollback to the last stable state.

Learn more →

4. ExecutionContract¶

After each observation, Identity OS emits an ExecutionContract — a deterministic constraint set that tells your agent what it can and can't do. Your agent reads this contract to make decisions.

Learn more →

How It Works (Developer Perspective)¶

Your Agent                    Identity OS
   |                              |
   |-- POST /process ------------>|
   |   (observation: what         |
   |    the agent just did)       |
   |                              |--- Engine evaluates
   |                              |--- Updates state
   |                              |--- Detects drift
   |                              |
   |<-- ExecutionContract --------|
   |   (allowed_actions,          |
   |    forbidden_actions,        |
   |    stress_level, etc.)       |
   |                              |
   |-- Agent uses contract ------>|
   |   to decide next action      |

That's it. One API call per turn. The engine handles all the complexity internally.