Ninja AI API

Ninja AI is the multi-agent AI system that powers autonomous remediation across the iTechSmart platform. It routes work across 40+ models — including NVIDIA Nemotron 49B, Google Gemini, OpenAI GPT-4o, Anthropic Claude, and Ninja-LLM 3.0 — through specialised agents (Scheduler, Researcher, Coder, Analyst). See the platform overview for the architecture.

Access model — read this first

There is no public standalone Ninja REST API today. The Ninja dashboard at ninja.itechsmart.dev sits behind the platform access gate (it 302-redirects to login). This page documents the ways Ninja’s capabilities are reachable programmatically right now — all verified live. When a dedicated Ninja API ships it will be documented here.

1. Via the API gateway

The gateway’s AI endpoints are backed by the same model routing:

Method	Path	Description
POST	/v1/classify	RAG-enhanced incident classification (Nemotron-backed). Seals a ProofLink receipt per call.
POST	/v1/validate	NeMo Guardrails validation of a proposed remediation action (4 rail types).

curl -X POST https://api.itechsmart.dev/v1/classify \
  -H "Content-Type: application/json" \
  -d '{"description": "disk usage above 95% on db-primary, writes failing"}'

Full request/response schemas, rate limits, and Python/JavaScript examples are on the UAIO API reference.

2. Via the MCP server

Agentic clients (Claude Code, Cursor, any MCP client) get the richest access. Two of the 66 tools on mcp.itechsmart.dev drive Ninja’s reasoning directly:

Tool	What it does
invoke_octoai_pipeline	Invoke the OctoAI 7-node cognitive pipeline with a free-form prompt.
dispatch_ag2_incident	Dispatch an IT incident to the AG2 6-agent GroupChat for autonomous diagnosis.

Connection guides (Claude Code, Cursor, raw SSE) are on the MCP Server page. Every MCP tool call requires a Bearer key and seals a ProofLink receipt.

3. Via the A2A protocol

The MCP server also speaks the Linux Foundation Agent-to-Agent (A2A) protocol at POST mcp.itechsmart.dev/a2a (agent card at GET /.well-known/agent.json, no auth for the card). Plain-text A2A messages route straight into the OctoAI reasoning pipeline — i.e. sending a text message to the agent is the lowest-friction way to talk to Ninja from another agent framework. A structured DataPart invokes a named skill instead. Details and a JSON-RPC example are in the MCP reference, A2A section.

Proof of every action

Whichever path you use, autonomous actions produce SHA-256 hash-chained ProofLink receipts on the public ledger at verify.itechsmart.dev — classification calls, pipeline invocations, and AG2 dispatches are all sealed and independently verifiable via the ProofLink API.

What is not public (yet)

• Direct REST access to individual Ninja agents (Scheduler, Researcher, Coder, Analyst) — gated behind the platform login.
• Model-selection / routing controls — routing is automatic today.
• The ninja.itechsmart.dev dashboard itself — requires a platform account.

If you need gated access for an evaluation, contact [email protected].