Avinish Kumar, home← All work

Case study · 2026

Agent Dashboard

Built it, AI Platform @ Leena AI

An AI operator agent inside the Leena AI admin console. Admins configure and debug bots by talking to it; it streams its replies, calls platform APIs as tools, and is exposed over A2A and MCP so other agents can drive the same workflows.

ClaudeNode.jsTypeScriptSSEA2AMCP

The problem

Configuring and debugging enterprise bots meant clicking through many admin screens and knowing where every setting lives.

The platform already had the APIs to do all of it. What was missing was an agent that could use them on an admin's behalf, safely, and show its work as it went.

One request through the Agent Dashboard

  1. admin consolean admin asks in plain language
  2. agent authagent token checked (JWT / ACL) before anything runs
    + external agentsarrive over A2A or MCP instead of the console
  3. agent (Claude)plans the change and picks tools
  4. tools → platform APIsreads and updates bot configuration
  5. SSE streamreply and tool progress stream back to the console

Key decisions

Tools are the platform's own APIs

The agent acts through the same APIs the product already exposes, so what it can do is bounded by what those APIs and the caller's permissions allow.

Stream everything over SSE

Admins see the reply and each tool step as it happens instead of waiting on a spinner, on the same streaming runtime the rest of the platform uses.

One agent, three front doors

The console, A2A and MCP all reach the same agent and workflows, so external agents get the capability without a second implementation to maintain.

Results

3
surfaces: console, A2A, MCP
500+
enterprises on the platform it runs in
Next case studyMulti-agent PR review pipeline →