DevTools · Multi-Agent

BlackBox AI

Multi-agent AI coding platform — parallel agents judged by an LLM, GCP execution sandboxes, LLM router, IDE integrations. Live since Feb 2026.

Tech Stack:TypeScriptNode.jsPythonGCPDocker

The Problem

Single-agent coding assistants hit quality ceilings: one model, one attempt, no arbiter. Teams running real production work need multiple agents racing on the same task, an independent judge picking the best diff, execution that happens in isolated sandboxes (not on a developer's laptop), and a billing / abuse-prevention layer that makes the whole thing viable as a hosted product. BlackBox wanted all of that in one shipping platform, not five disconnected components.

The Solution

Apex36 shipped a five-component cohesive multi-agent platform, live since Feb 2026: the BlackBox CLI (@blackbox_ai/blackbox-cli on npm), a multi-agent runner dispatching opencode, qwen-code, and other agents in parallel on the same task, a judge LLM that compares git diffs and picks the winner, GCP-hosted isolated Docker sandboxes on Amazon Linux 2023 with per-sandbox billing, a Python LLM router handling API-key management, auto-topup billing, and abuse prevention, plus Open Interpreter VS Code server fork and RoboCoder IDE integrations.

Features

BlackBox CLI

Command-line coding agent published as @blackbox_ai/blackbox-cli on npm, the entry point to the entire platform.

Multi-agent runner + judge

Dispatches multiple coding agents (opencode, qwen-code, and others) in parallel on the same task; a judge LLM compares their git diffs and selects the winning implementation.

Remote code execution

GCP-hosted isolated Docker sandboxes (Amazon Linux 2023) with a command-palette UI and GCP-based billing for every sandbox minute.

Python LLM router

Routes every LLM call through API-key management, auto-topup billing, and abuse prevention, with an admin panel for operators.

VS Code integrations

Open Interpreter VS Code server fork plus the RoboCoder in-IDE agent extension, the same platform reachable from inside VS Code.

Results / Impact

Live since Feb 2026

real multi-agent runs against real tasks logged in production.

@blackbox_ai/blackbox-cli

published on npm as the developer entry point.

Parallel agents + judge LLM

selection over git diffs.

GCP-hosted sandboxes

isolated Docker containers on Amazon Linux 2023, billed per sandbox-minute.

FAQ

Several coding agents (opencode, qwen-code, and others) run in parallel on the same task. Each produces a git diff. A judge LLM then compares those diffs and selects the winning implementation. The user sees the chosen result, not the parallel work behind it.

Yes. The platform has been in active production since Feb 2026 and the `blackbox` CLI is published on npm as @blackbox_ai/blackbox-cli.

No. Execution runs in GCP-hosted isolated Docker sandboxes (Amazon Linux 2023) with per-sandbox billing and abuse prevention. Developers interact via CLI or VS Code integrations; the work happens remotely.

Apex36, starting as an SWE trial in Feb 2026 that converted into an active paid engagement covering CLI, multi-agent runner + judge, remote execution, LLM router, and VS Code integrations.

Ready to build something impactful?

Let's discuss your project and how we can help you ship faster and smarter.

Book a Free Strategy Call