Enterprise Agent PlatformSafe, observable, governable Multi-Agent Control Plane

Data · Knowledge · Ops · Supervisor — with RBAC, Tool Policy, Human Approval, Run/Trace, Eval and enterprise operations built in.

Open Live Demo

Interview Guide

GitHub

Multi-Agent by domain

Data, Knowledge and Ops keep their own tool and security boundaries. Supervisor orchestrates only when a real cross-domain investigation needs it.

Security outside the prompt

RBAC, Tool Policy, SQL Guard, allowlists and target-bound single-use approvals enforce capabilities independently from model output.

Observable execution

Request ID, Correlation ID, Run, Trace, structured logs, P50/P95 metrics and Prometheus make failures diagnosable end to end.

Regression-ready

Deterministic evals and isolated Fake-LLM regression cases run in CI without external model keys or production databases.

Resource-oriented console

Agents, Knowledge, Data Sources, Runs, Approvals, Integrations and Credentials are modeled as first-class platform resources.

Deliberately simple infrastructure

SQLite and in-process components keep the demo self-contained, while interfaces keep PostgreSQL, pgvector, OIDC and shared state as clear upgrade paths.

What this project demonstrates

Enterprise Agent Platform focuses on the engineering layer around LLM capabilities: how agents are constrained, audited, observed, evaluated and safely allowed to act.

01Get started

Run the platform locally and understand the fastest demo path.

02Architecture

See the platform boundaries, multi-agent flow and approval model.

03Demo guide

Use the recommended interview walkthrough from login to trace.

04Interview handbook

Project pitch, design trade-offs and 65 high-frequency questions.

Core execution model

text

Web Console / API
        │
        ▼
Authentication + RBAC
        │
        ├──────── Data Agent
        ├──────── Knowledge Agent
        ├──────── Ops Agent
        └──────── Supervisor
                     │
                     ├── Ops evidence
                     ├── Knowledge evidence
                     └── Data evidence

All execution is governed by:

Tool Policy → Human Approval → Run / Trace → Eval → Metrics

Live demo

Open agent.majhoon.site and use the demo credentials: