AgentOps — AI Agent Governance Platform

Jira meets Workday for your AI workforce. AgentOps is a production-grade B2B platform for governing, monitoring, and controlling fleets of autonomous AI agents — with human-in-the-loop approvals, immutable audit logs, cost governance, and a conversational Ops Copilot.

Live demo: https://v0-agentops-platform-build.vercel.app/ Alternative demo (preview): https://v0-agentops-platform-build-ozdjtqp7s-abhay-zangir.vercel.app/

Demo login: ops@company.ai / AgentOps2024! (or click "Fill demo credentials" on the login screen)

---

Overview

Enterprises deploy autonomous agents that execute actions with real cost and safety implications. AgentOps provides the missing governance layer: permissions, approvals, telemetry, cost controls, and compliance summaries — all in one product.

Key Features

• Command Center: real-time telemetry & health
• Agent Registry: hierarchical agent model with recursive permission inheritance
• Human-in-the-loop Approvals: approve/reject/modify high-risk actions
• Sentinel AI: automated risk analysis (AWS Bedrock)
• Cost Intelligence: per-agent budgets, alerts, and spend freeze
• Immutable Audit Log: exportable CSV/JSON and AI summaries
• Ops Copilot: conversational assistant grounded in live fleet data
• Billing & Plans: subscription management (Razorpay)

Architecture

Mermaid diagram (rendered on GitHub):

flowchart LR
  subgraph Frontend
    A[Next.js App (Vercel)]
  end

  subgraph Backend
    B[Serverless API Routes]
    C[Aurora PostgreSQL] -->|ACID registry, approvals, audit log| B
    D[DynamoDB (Activity Stream)] -->|SSE / poll| B
    E[AWS Bedrock] -->|Risk analysis & Copilot| B
    F[Razorpay] -->|Billing webhooks| B
  end

  A -->|HTTP / Web| B
  B --> C
  B --> D
  B --> E
  B --> F
  D -->|SSE| A
  E -->|LLM responses| A

  classDef cloud fill:#f9f,stroke:#333,stroke-width:1px;
  class C,D,E,F cloud;

Loading

High-level data flow

1. Agents emit actions -> persisted to DynamoDB (fast, high-throughput).
2. Serverless API persists governance decisions & metadata to Aurora (system of record).
3. Approvals created in Aurora; Sentinel AI (Bedrock) analyzes pending approvals and returns structured risk reasoning.
4. Frontend subscribes to SSE stream for live activity and uses SWR for data fetching.

Tech Stack

• Next.js (App Router) + React + TypeScript
• TailwindCSS + shadcn/ui
• Aurora PostgreSQL Serverless v2 (RDS)
• DynamoDB (activity stream)
• AWS Bedrock for LLMs
• Vercel for hosting
• Razorpay for billing

Live Links

• Production preview / demo: https://v0-agentops-platform-build.vercel.app/
• v0 app session examples: https://v0.app/chat/zvct1uUcMy3

Quickstart (Local)

Prerequisites:

• Node 18+ and pnpm
• AWS account with RDS, DynamoDB, and Bedrock access (or use mock/stubbed services)
• Razorpay account (for billing features)

Steps:

pnpm install
cp .env.example .env.development.local
# Fill environment variables
pnpm dev # http://localhost:3000

Initialize DB (local dev only — requires DB credentials):

curl -X POST http://localhost:3000/api/migrate
curl -X POST http://localhost:3000/api/seed

Environment Variables (.env.example)

Copy this file to .env.development.local or set the values in Vercel project settings.

# Aurora PostgreSQL
PGHOST=your-db-host
PGDATABASE=agentops
PGUSER=db_user
# If using IAM auth: leave PGPASSWORD blank and configure AWS_ROLE_ARN
PGPASSWORD=
AWS_REGION=us-east-1
AWS_ROLE_ARN=arn:aws:iam::123456789012:role/your-role

# DynamoDB
DYNAMODB_TABLE_NAME=agentops-activity
DYNAMODB_TABLE_PARTITION_KEY=pk

# AWS Bedrock (AI)
BEDROCK_AWS_REGION=us-east-1
BEDROCK_AWS_ACCESS_KEY_ID=
BEDROCK_AWS_SECRET_ACCESS_KEY=
BEDROCK_MODEL_ID=qwen-3

# Auth
AUTH_SECRET=replace_with_a_secure_random_value

# Razorpay
RAZORPAY_KEY_ID=
RAZORPAY_KEY_SECRET=
RAZORPAY_GROWTH_PLAN_ID=
NEXT_PUBLIC_RAZORPAY_KEY_ID=

# Optional local dev toggles
NEXT_PUBLIC_USE_DEMO_SEEDS=true

Feature Walkthrough

• Command Center: shows aggregated agent health, spend trend charts, and active alerts.
• Agent Registry: create an agent, assign parent agent, set capability scopes, and define budget caps.
• Approvals: any action flagged as high-risk appears in the Approvals inbox — view Sentinel AI summary, then Approve/Reject/Modify.
• Audit Log: searchable, filterable, and exportable — use the AI summary endpoint for compliance-ready narratives.

Code Quality & Formatting

• Internal conventions: TypeScript strict mode, ESLint, Prettier.
• Formatting config included: .prettierrc
• Please run pnpm lint and pnpm format before submitting PRs.

Contributing

See CONTRIBUTING.md for guidelines (if present). Open issues for bugs or feature requests.

Contact

Maintained by @abhayzangir1 — raise issues or PRs in this repository.