StreamOps CommandMesh

A governed multi-agent execution system for enterprise QA incident response built on the Airia platform.

What is StreamOps CommandMesh?

StreamOps CommandMesh is an 8-agent AI pipeline that autonomously handles QA incident response for enterprise streaming platforms. When a CI/CD failure or API degradation is detected, the pipeline triggers automatically:

Intake → Analyst → Risk → Planner → Approval (HITL) → Action → Document → Audit

No human needs to manually triage, escalate, notify, or document — the agents do it all, with a human-in-the-loop approval gate before any remediation actions are executed.

The Problem

Enterprise QA teams in media/streaming face:

• CI/CD failures that take hours to triage and escalate
• Manual Jira ticket creation, Slack notifications, and incident reports
• No audit trail for compliance
• Approval bottlenecks causing delayed remediation

Architecture

flowchart TD
    trigger["🔔 Incident Trigger<br/><i>CI/CD · API · Jira</i>"]
    intake["1 · Intake Agent<br/>Classify type, severity, system"]
    analyst["2 · Analyst Agent<br/>Enrich: root cause, SLA, history"]
    risk["3 · Risk Agent<br/>Score: LOW / MED / HIGH / CRIT"]
    planner["4 · Planner Agent<br/>Generate remediation plan"]
    approval{{"5 · Approval Agent<br/>⚠️ HITL Gate<br/>APPROVE / REJECT"}}
    action["6 · Action Agent<br/>Jira + Slack + CI/CD Rollback"]
    document["7 · Document Agent<br/>Incident Report + Compliance"]
    audit["8 · Audit Agent<br/>Immutable Log + Signature"]

    trigger --> intake
    intake --> analyst
    analyst --> risk
    risk --> planner
    planner --> approval
    approval -- "✅ APPROVED" --> action
    action --> document
    action --> audit

    style trigger fill:#1e3a5f,stroke:#4a90d9,color:#ffffff,stroke-width:2px
    style intake fill:#1a5276,stroke:#5dade2,color:#ffffff,stroke-width:2px
    style analyst fill:#1a5276,stroke:#5dade2,color:#ffffff,stroke-width:2px
    style risk fill:#7b241c,stroke:#e74c3c,color:#ffffff,stroke-width:2px
    style planner fill:#1a5276,stroke:#5dade2,color:#ffffff,stroke-width:2px
    style approval fill:#7d6608,stroke:#f1c40f,color:#ffffff,stroke-width:3px
    style action fill:#0e6655,stroke:#1abc9c,color:#ffffff,stroke-width:2px
    style document fill:#1a5276,stroke:#5dade2,color:#ffffff,stroke-width:2px
    style audit fill:#4a235a,stroke:#af7ac5,color:#ffffff,stroke-width:2px

Loading

Stages 7 and 8 are called as nested agents by the Action Agent — keeping orchestration clean and each agent stateless.

Pipeline Flow

flowchart LR
    A["1 · Intake<br/><b>JSON</b>"]
    B["2 · Analyst<br/><b>JSON</b>"]
    C["3 · Risk<br/><b>Score</b>"]
    D["4 · Planner<br/><b>Plan</b>"]
    E{{"5 · Approval<br/><b>HITL Gate</b>"}}
    F["6 · Action<br/><b>Execute</b>"]
    G["7 · Document<br/><b>Report</b>"]
    H["8 · Audit<br/><b>Log</b>"]

    A --> B --> C --> D --> E -- "✅" --> F --> G --> H

    style A fill:#1a5276,stroke:#5dade2,color:#ffffff,stroke-width:2px
    style B fill:#1a5276,stroke:#5dade2,color:#ffffff,stroke-width:2px
    style C fill:#7b241c,stroke:#e74c3c,color:#ffffff,stroke-width:2px
    style D fill:#1a5276,stroke:#5dade2,color:#ffffff,stroke-width:2px
    style E fill:#7d6608,stroke:#f1c40f,color:#ffffff,stroke-width:3px
    style F fill:#0e6655,stroke:#1abc9c,color:#ffffff,stroke-width:2px
    style G fill:#1a5276,stroke:#5dade2,color:#ffffff,stroke-width:2px
    style H fill:#4a235a,stroke:#af7ac5,color:#ffffff,stroke-width:2px

Loading

The Solution

StreamOps CommandMesh compresses incident response from hours to seconds:

Stage	Agent	Output
1	Intake Agent	Classifies and routes the incident
2	Analyst Agent	Enriches with SLA data and root cause
3	Risk Agent	Scores risk (LOW/MEDIUM/HIGH/CRITICAL)
4	Planner Agent	Generates structured remediation plan
5	Approval Agent	HITL gate — human approves/rejects
6	Action Agent	Executes Jira, Slack, CI/CD rollback
7	Document Agent	Generates incident report and compliance docs
8	Audit Agent	Produces immutable audit log with signature

Try It Live

Note: A free Airia account is required to run the agents. Signup takes under 60 seconds at airia.ai.

• Try StreamOps Intake Agent — Start here, pipeline entry point
• Try StreamOps Action Agent — Full execution: Jira + Slack + Rollback

Tech Stack

• Platform: Airia Agent Studio
• Model: GPT-4o (OpenAI)
• Integrations: Slack, Microsoft Teams, Jira (mocked in demo)
• Output format: Structured JSON + natural language dual-output
• Orchestration: Nested agent calls, HITL approval gate

Golden Path Demo

See prompts/golden-path-demo.md for the exact input prompt and expected outputs used in testing.

Agent Specifications

See the agents/ folder for full prompt design, input/output schema, and purpose for each of the 8 agents.

Architecture Deep Dive

See architecture.md for design decisions: why 8 agents, why HITL at Stage 5, nested agent patterns, and dual JSON + natural language output.

Built By

Partha Samal — QA Engineer / AI Automation Specialist Airia AI Agents Hackathon 2026