Compute API · Run inference on the network from your code

Compute API

Run Inference on the Network From Your Own Code

A developer API over the same operator nodes ParalleliX AI uses. Point an OpenAI-compatible client at it, fan a batch of prompts across the network in parallel, or wire it into Claude with the MCP connector. Calls spend $PRLX credits.

npm

$npxparallelix-mcp

Create an API key Read the docs

// three doors, one billing rail

OpenAI-compatible

POST /v1/chat/completions in the OpenAI format. Point Cursor, an SDK, n8n, or any existing client at ParalleliX by changing one base URL.

Parallel batch

parallel-native

POST /v1/batch with an array of prompts. They fan out across the online nodes at once, each returning its own result and Proof-of-Execution. One submission, N sub-tasks in parallel.

MCP connector

npx parallelix-mcp gives Claude a parallel_map tool. Your agent orchestrates and reasons; the open-source fleet runs the bulk parallel sub-tasks.

parallelix-mcp

Bulk parallel work, from inside Claude

The connector adds a parallel_map tool to Claude Code and Claude Desktop. Hand it a list (500 reviews to classify, 40 files to summarize) and a single instruction; it fans the items across the network at once and returns one result per item, each with a Proof-of-Execution. Your frontier model stops burning tokens on the cheap embarrassingly-parallel parts.

npm Source

// claude_desktop_config.jsonmcp

{
  "mcpServers": {
    "parallelix": {
      "command": "npx",
      "args": ["-y", "parallelix-mcp"],
      "env": { "PARALLELIX_API_KEY": "pk_live_your_key" }
    }
  }
}

// api.parallelix.io/v1

# api.parallelix.io
POST /v1/chat/completions   OpenAI-compatible completion
POST /v1/batch              fan prompts out across the network
GET  /v1/batch/{id}         per-item result + Proof-of-Execution
GET  /v1/models             models the network serves now
GET  /v1/usage              requests, credits spent, balance

# Authorization: Bearer pk_live_...   spends $PRLX credits
# 85% of every paid call settles to the serving operators

One key. One billing rail.

One key

Create an API key in the app under Developers. It points at your wallet's $PRLX credit balance.

Metered per call

Each call is metered off-chain against that balance, the same balance ParalleliX AI uses. No per-call gas.

85% to operators

Every paid call settles 85% on-chain to the operators whose nodes served it. Each result carries a Proof-of-Execution.

The network runs open-source models (currently 7B-class). The Compute API is a cheap parallel executor for bulk independent sub-tasks, not a frontier model. Capacity is the set of nodes online at request time, and GET /v1/models reports it.

Get a key, make your first call

Create an API key Full documentation