A·B
v2

Agent Black

Multi-Agent AI Research Platform

A multi-agent system where a Control Panel orchestrates three specialized research agents to solve AI/ML/CV/NLP queries. Each agent is an independent FastAPI service with its own MCP tool server and A2A endpoint — delivering structured, end-to-end research workflows.

Agents
3

CV · NLP · ML

MCP Tools
39

13 per agent

Pipeline Steps
5

End-to-end

LLM Providers
3

Gemini · OpenAI · Anthropic

Key Features

What makes Agent Black powerful.

Research-Relevance Gate

Smart query validation rejects non-research questions with helpful suggestions and supported topics.

LLM-Driven Orchestration

Agent selection, task decomposition, and result aggregation powered by multi-provider LLMs.

39 MCP Tools

13 tools per agent covering paper search, dataset discovery, model recommendation, and more.

A2A Protocol

Agents communicate seamlessly via JSON-RPC 2.0 with standardized agent cards and task routing.

SSE Streaming

Real-time task progress streaming so you never lose sight of what's happening.

Structured Reports

Unified reports with literature review, datasets, models, evaluation plans, and PDF export.

Orchestrator Pipeline

Every query flows through 5 intelligent steps.

0

Relevance Gate

Validates research intent

1

Agent Selection

LLM picks the right agents

2

Task Decomposition

Breaks into sub-tasks

3

Concurrent Dispatch

Parallel agent execution

4

Result Aggregation

LLM synthesizes output

Specialist Agents

Each agent is a standalone FastAPI service with its own MCP tools and A2A endpoint.

CV Research Agent

Port 8001 · Computer Vision

Image ClassificationObject DetectionSegmentationVision TransformersMedical ImagingVideo Analytics

NLP Solution Agent

Port 8002 · NLP / Solution Architecture

LLMs & RAGPrompt EngineeringText ClassificationSummarizationConversational AIInformation Extraction

ML Experiment Agent

Port 8003 · Machine Learning Experiments

Classical MLFeature EngineeringHyperparameter TuningTime SeriesExperiment DesignModel Explainability

Team

The people behind Agent Black.

Tech Stack

Technologies powering the platform.

Frontend
React 19
UI library
TanStack Start
SSR + file-based routing
Tailwind CSS v4
Utility-first CSS
shadcn/ui
Component primitives
Zustand
Persisted state
TanStack Query
Data fetching + caching
Mermaid.js
Diagram rendering
Vite 7
Build tool
Backend
FastAPI
Async Python API
Uvicorn
ASGI server
Pydantic v2
Data validation
SQLite
Config & history (WAL mode)
HTTPX
Async HTTP client
Agents & Protocols
MCP (JSON-RPC 2.0)
39 tools across 3 agents
A2A Protocol
Agent-to-Agent communication
SSE Streaming
Real-time task progress
Academic APIs
CrossRef + Semantic Scholar + arXiv
LLM Providers
Google Gemini
gemini-1.5-flash default
OpenAI
gpt-4o
Anthropic
claude-3-5-sonnet
Auto-Retry
Automatic retry on failure

Quick Start

1pip install -r requirements.txt
2python start.py
3cd ui && bun install && bun run dev

Service Ports

Control Panel:8000
CV Agent:8001
NLP Agent:8002
ML Agent:8003
Frontend:8080
A·B

Agent Black · Built with FastAPI, React, and Tailwind CSS

© 2026 Agent Black Contributors · MIT License