NiFiPilot Agents

AI agents that control Apache NiFi via natural language using the NiFiPilot MCP server.

---

What is this

NiFiPilot Agents is an orchestration layer that lets any AI model control Apache NiFi flows via the NiFiPilot MCP server.

• Provider-agnostic — works with Claude, GPT-4, Gemini, Llama, DeepSeek and any LiteLLM-compatible model
• Multi-agent — specialized sub-agents for monitoring, building and operations
• Skill-based — agents follow the NiFiPilot Agent Skill for safe, consistent behavior
• Free to develop — use Ollama locally with no API costs

---

Architecture

User request
     ↓
NiFi Supervisor Agent (orchestrator)
     ↓
┌────────────────────────────────────┐
│  Monitor Agent  │  Builder Agent   │
│  (read-only)    │  (creates flows) │
├────────────────────────────────────┤
│         Ops Agent                  │
│  (start/stop/control)              │
└────────────────────────────────────┘
     ↓
NiFiPilot MCP Server (pip install nifipilot)
     ↓
Apache NiFi 2.x REST API

---

Quick start

1. Install NiFiPilot MCP server

pip install nifipilot

Start the MCP server (or use Docker Compose from the NiFiPilot repo):

NIFI_URL=https://localhost:8443 \
NIFI_USERNAME=admin \
NIFI_PASSWORD=your-password \
MCP_MODE=full \
nifi-mcp

2. Install NiFiPilot Agents

git clone https://github.com/Giojacke/nifipilot-agents
cd nifipilot-agents
pip install -e .

3. Configure your AI provider

cp .env.example .env

Edit .env and choose your provider:

# Option A — Anthropic Claude (recommended)
LITELLM_MODEL=claude-sonnet-4-6
ANTHROPIC_API_KEY=your-key

# Option B — OpenAI
# LITELLM_MODEL=gpt-4o
# OPENAI_API_KEY=your-key

# Option C — Ollama (free, local, no API key needed)
# LITELLM_MODEL=ollama/llama3.2

4. Run the monitor agent

python agents/sub_agents/monitor_agent.py

---

Agents

Monitor Agent (available now)

Read-only agent specialized in NiFi health checks and diagnostics.

python agents/sub_agents/monitor_agent.py

Capabilities:

• System health check (heap, CPU, uptime)
• Flow status inspection
• Queue depth monitoring
• Issue detection and reporting

Builder Agent (coming soon)

Creates NiFi flows from natural language descriptions.

Ops Agent (coming soon)

Controls start/stop/control operations on NiFi flows.

Supervisor Agent (coming soon)

Orchestrates sub-agents for complex multi-step tasks.

---

Supported AI providers

Provider	Model example	Cost
Anthropic	claude-sonnet-4-6	~$3/M tokens
Anthropic	claude-haiku-4-5	~$0.25/M tokens
OpenAI	gpt-4o	~$5/M tokens
Google	gemini/gemini-pro	Free tier available
Ollama	ollama/llama3.2	Free (local)
DeepSeek	deepseek/deepseek-chat	~$0.14/M tokens

---

Using the Agent Skill

The skills/nifipilot-agent.md file contains the instruction set that guides agents to use NiFiPilot tools safely.

To use it with any AI:

Claude Code:

Read skills/nifipilot-agent.md and use it as your instruction set.
Then check the current state of NiFi.

VS Code / Cursor: Pin skills/nifipilot-agent.md to the chat context before starting a NiFi session.

---

Cost control

For development, use Ollama (free):

# Install Ollama: https://ollama.com/download
ollama pull llama3.2

# In .env:
LITELLM_MODEL=ollama/llama3.2

For production, set a spending limit:

• Anthropic: https://console.anthropic.com/settings/limits
• OpenAI: https://platform.openai.com/settings/organization/limits

---

Requirements

• Python 3.11+
• NiFiPilot MCP server running (pip install nifipilot)
• Apache NiFi 2.x instance
• One of: Anthropic API key, OpenAI API key, or Ollama installed

---

Related projects

• NiFiPilot MCP Server — the MCP server this agent uses
• NiFiPilot on PyPI — pip install nifipilot
• NiFiPilot VS Code Extension

---

License

MIT

---

Author

Built by CodHector 🤙