Skip to main content
ParalleliX Docs

// Chapter 18 · Products

ParalleliX CODE

A terminal coding agent whose inference runs on the ParalleliX decentralized GPU network.

12 min11 sectionsProducts

A coding agent backed by decentralized GPU.

// parallelix code · terminal repl · open-source models · sha-256 poe · $prlx credits

// What it does

Install once. Authenticate with a pk_live_ key. Run parallelix-code in your project. Type a task. The agent reads your files, writes changes, runs commands, and returns a Proof-of-Execution after each network call.

Every inference step calls api.parallelix.io/v1/chat/completions is the same OpenAI-compatible endpoint the rest of the ParalleliX AI surface uses. A network node serves the model on its local GPU, the coordinator verifies the result, and a SHA-256 hash commitment is returned with each response. The agent runs on open-source models: lower quality than frontier services, fully auditable execution path.

// install + auth + first task
$ curl -fsSL https://parallelix.io/install-code.sh | sh
✓ parallelix-code installed

$ parallelix-code auth pk_live_…
✓ authenticated

$ parallelix-code
parallelix-code v1.0.0 · decentralized coding agent

> refactor auth middleware to use JWT
reading src/middleware/auth.ts...
⬡ node 12  sha256:a3f9c2…  0.0420 $PRLX
writing src/middleware/auth.ts...
Done.
Get an API keyDeposit credits

What it is and what it is not

// open-source models · lower quality than frontier · verifiable execution

IS

A terminal coding agent

  • Reads, writes, and runs code in your project directory
  • Driven by open-source models on decentralized GPU nodes
  • Every call returns a SHA-256 Proof-of-Execution
  • Billed in $PRLX credits. Deposit once, no per-call gas
  • OpenAI-compatible API at api.parallelix.io/v1
IS NOT

A frontier model service

  • Does not match GPT-4 or Claude quality; open-source models only
  • Not a browser extension or GUI
  • Not a code completion plugin (it is a task-level REPL)
  • Not backed by centralized cloud GPU; network nodes serve it
  • Not free: each inference call costs $PRLX credits

Three steps to a running agent

// install · auth · repl · ready in under two minutes

// Quick start sequence

  1. Install

    curl -fsSL https://parallelix.io/install-code.sh | sh

    Places parallelix-code on your PATH. macOS and Linux x86_64. Windows requires WSL2.

  2. Authenticate

    parallelix-code auth pk_live_<your-key>

    Get a pk_live_ key at ai.parallelix.io/developers. Written to ~/.parallelix-code/config.json.

  3. Start the REPL

    parallelix-code

    Opens the interactive REPL. Type any coding task in plain language. /clear resets context. exit to quit.

How a task executes

// repl → api → node → poe → agent acts · 30-turn max per task

// Execution path · 5 stepsmax 30 turns

  1. // 01

    REPL receives task

    You type a task. The agent reads your project files to build context.

  2. // 02

    API call to api.parallelix.io

    The agent posts to api.parallelix.io/v1/chat/completions, an OpenAI-compatible endpoint routed to the network.

  3. // 03

    Node serves inference

    A network node runs the request on its local GPU through Ollama and returns a signed result.

  4. // 04

    PoE printed

    The coordinator verifies the result and returns the SHA-256 Proof-of-Execution. The REPL prints: node ID, hash, cost.

  5. // 05

    Agent acts on the result

    The model output is parsed for tool calls (read_file, write_file, bash, patch, etc.). The loop repeats up to 30 turns.

// Guide sections · 11 entries