API

Embed GPU inference in your stack.

Versioned REST + GraphQL. NIM-compatible endpoints. Streaming SSE for live inference output. Type-safe SDKs in TypeScript and Python.

Trigger a workflow

POST to /v1/workflows/:slug/runs. Stream output via SSE.

POST/v1/workflows/:slug/runs
GET/v1/runs/:id
GET/v1/runs/:id/events (SSE)
POST/v1/runs/:id/approve
GET/v1/policies
GET/v1/connectors

curl

curl -X POST https://api.mandarum.com/v1/workflows/  nim-inference-pipeline/runs   -H "Authorization: Bearer $MANDARUM_KEY"   -H "Content-Type: application/json"   -d '{
    "input": { "model": "meta/llama-3.1-70b-instruct", "prompt": "Summarize the Q4 report" },
    "stream": true
  }'

TypeScript SDK

Type-safe, edge-compatible. npm i @mandarum/sdk

Python SDK

Async-first, batteries-included. pip install mandarum

MCP servers

Expose your tools via MCP and they appear in every workflow.