DecisionOS

DecisionOS transforms noisy operational data (tickets, metrics, market signals) into ranked, explainable decisions. It is designed for high-throughput enterprise environments where auditability and specific confidence calibration are required.

🖥️ Web Demo (Recommended)

Try the full decision pipeline in 3 minutes via the browser-based demo.

1. Start the localized server:

   cd src
   poetry run uvicorn decisionos.main:app --reload

2. Open the Dashboard:

   • Navigate to http://localhost:8000/demo

3. Walkthrough Flow:

   • Observe State: See the "Live Feed" waiting for input.
   • Inject Incident: Click "Inject Synthetic Incident". This simulates a Latency Spike in web-tier-04.
   • Watch Reasoning: As the decision processes, expand the card to see agents (Signal, Decision, Critic) debating in real-time.
   • Review Output: Note the final "DECLARE_SEV1" decision and calibrated confidence score.
   • Reset: Click "Reset Demo" to clear the state and record another take.

🏗️ System Architecture

graph TD
    Client[API Client] -->|POST /ingest| API[FastAPI Gateway]
    Client -->|POST /generate| API

    subgraph Core Infrastructure
        API -->|Enqueue| Q[Redis Queue]
        Q -->|Consume| Worker[Celery Worker]
        Worker -->|Read/Write| DB[(PostgreSQL)]
    end

    subgraph Decision Engine
        Worker -->|1. Normalize| Norm[Normalizer]
        Norm -->|2. Extract| Signals[Signal Engine]
        Signals -->|3. Reason| Agents[Multi-Agent System]
        Agents -->|4. Score| Scoring[Scoring Engine]
        Scoring -->|5. Explain| Explain[Explainer]
        Explain -->|6. Govern| Govern[Governance]
    end

    Govern -->|Persist| DB

Loading

🚀 Key Features

• Event-Driven Ingestion: Asynchronous processing pipeline using Celery and Redis.
• Deterministic Normalization: Strict Pydantic validators ensure "Garbage-In, Garbage-Out" protection.
• Multi-Agent Reasoning:
  • SignalAgent: Pattern recognition.
  • DecisionAgent: Strategy proposal.
  • CriticAgent: Adversarial validation (Red Teaming).
  • SupervisorAgent: Final synthesis.
• Calibrated Scoring: Risk-adjusted confidence intervals (not just single-point estimates).
• Regulatory Audit Trail: Full lineage tracing from Decision -> Agent Thoughts -> Raw Inputs.

Primary Use Case

Scenario: Ops Team Prioritizing Incidents

User Persona: Site Reliability Engineer (SRE) Lead

Critical Decision: "Which of the 200+ weekly alerts indicate a genuinely critical system failure requiring immediate mobilization, versus noise that can be scheduled for later?"

📚 Documentation

• Deployment Guide
• Operational Runbook
• API Specification (Localhost default)

Tenex Build First

• Faster Decisions: Reduces Mean Time To Decision (MTTD) by automating the correlation of disparate signals (logs, tickets, deployments). Where humans spend hours triangulating root causes across dashboards, the Multi-Agent engine synthesizes evidence in seconds.
• New Opportunities: Surfaces "invisible" correlation risks. For example, identifying that a minor database CPU spike (Infrastructure) consistently precedes "Checkout Failed" tickets (Support) by 15 minutes, allowing pre-emptive rollback before revenue impact.
• Measurable Impact: Moves beyond vague "better insights" to concrete metrics. Every decision generates estimated_time_saved_minutes and estimated_risk_reduction_score, providing an auditable ROI trail for every automated action.

---

Handover Status: Core infrastructure, pipeline logic, and governance frameworks are complete. Ready for ML model injection.