Free pilot — no card required. There are 20 account places. Apply for a place, verify your email, and create your account without manual approval. A place is held only when your invitation is used to create an account. Verification links expire after 24 hours and unused invitations after 7 days; neither reserves a place. When all places are claimed, the pilot is full.

Pilot members receive up to four managed AI provider attempts per UTC day, resetting at midnight UTC. One action may use several attempts; retries, history shortening, failed and cancelled provider attempts count. Existing members and administrators retain their current daily limits.

Managed AI accepts bounded text requests up to 8,000 input tokens and caps output at 2,000 tokens. Conservative size checks can reject shorter requests. Images, PDFs and tool calls are unsupported during the pilot; oversized or unsupported requests are rejected before dispatch.

FluxBench pays for managed AI. Everyone shares a $40 monthly reservation budget, resetting on the first day of each month at midnight UTC. Managed AI pauses when there is insufficient budget for an attempt, even if daily allowance remains. Unconfirmed costs stay reserved. There are no automatic charges or paid upgrades. Personal API keys use separate provider billing and are unaffected by this shared budget.

Cloud compilation, synthesis, simulation and extraction share ten operations per minute and two simultaneous operations per account. Cloud compilation depends on an externally hosted compiler and its availability.

The Windows 10/11 x64 local compiler preview requires the companion, board packages and required libraries, with no AI subscription required. It supports a bounded set of Arduino targets; FPGA synthesis is outside the local preview. The website and account/project services still require internet. Compiler setup and supported targets.

Saved project payloads are limited to 2 MB for files/designs and 400,000 code characters. Unlimited storage is not promised.

← Software support

Looking for local firmware builds? Download the separate compiler companion.

Choose your AI. Connect your computer.

In the workbench, select the AI: provider ▾ button to open AI & connections. Pick a provider, complete its setup, then select Save settings. The button shows your saved choice. Changing a selection without saving leaves your current provider in place.

Available AI connections
OptionWhat you need
Codex on your computerCodex signed in with ChatGPT and the FluxBench companion running; shares your subscription limits.
Claude, OpenAI or Gemini APIYour provider’s API key and separate API billing; no companion required.
Managed DeepSeek / LlamaNo personal API key; uses FluxBench’s managed AI allowance.

The Codex companion

Use your own ChatGPT-backed Codex account for FluxBench AI. Each FluxBench account has a separate connection. Your computer must remain awake and the companion must be running. Codex requests share your subscription limits; they never switch automatically to paid API credits.

Windows companion · preview

Windows 10/11, 64-bit. The setup app includes its own Node.js runtime, installs for your Windows user and adds a Start menu shortcut. No administrator access or terminal commands are needed to run the companion. Codex itself must already be installed and signed in.

Download Windows setup (.exe)

This preview is unsigned, so Windows may show an unknown-publisher warning. A signed release and additional-user setup testing remain pending. Download checksum · Version and build details

  1. Install Codex CLI if needed and sign in with ChatGPT.
  2. Open the downloaded setup app. Select Check Codex. If it is not found, use Browse to select the actual codex.exe. Choose an available model and name this computer.
  3. In FluxBench, choose AI & connections → Codex → Create setup token. Paste the token into the setup app and select Pair computer. This explicitly replaces an existing paired connection.
  4. Select Start. In FluxBench, select Refresh connection, then Save settings. The header will show Codex and its connection status.

Minimize the companion to keep it running in the system tray; double-click its tray icon to reopen. Stop pauses the connection. Closing the app also stops it. Reopen from Start → FluxBench Companion and select Start. Automatic startup is not installed.

Already connected? You can keep your current companion. Downloading or opening this app does not replace it; pairing with a new token does.

Manual setup · source ZIP

  1. Have Node.js 22 or newer and a local Codex CLI signed in with ChatGPT. You will need the absolute path to the Codex executable. On Windows, use the actual codex.exe, not a command shim.
  2. Sign in to FluxBench, download the companion, and extract it into a folder you trust. Open a terminal in that folder.
  3. Run node paired.mjs setup. Enter your local Codex executable path, choose a model from the available list, and name this computer.
  4. In FluxBench’s AI & connections, choose Codex and select Create setup token. Paste that token only into your local companion’s setup prompt. It expires after ten minutes and can be used once. Using a new token replaces your existing paired computer.
  5. Run node paired.mjs start. Leave the terminal open. Back in AI settings, select Refresh connection, then Save settings to use Codex.

Stopping, replacing and troubleshooting

Press Ctrl+C in the companion terminal to stop. Restart with the same start command. Select Disconnect companion in AI settings to revoke it and cancel pending work. To replace the computer, create a new token and run setup on the new computer. The original owner’s installed companion continues to work until it is explicitly replaced.

Only one request runs at a time. Replies appear after completion. Offline, interrupted, expired or busy requests fail visibly. Check your local Codex sign-in and usage allowance before retrying; failed requests may have used subscription capacity. Idle pickup can take up to fifteen seconds.

To remove the Windows app: close it, remove the FluxBench Companion Start menu shortcut and the version folder under %LOCALAPPDATA%\FluxBench.ai2\desktop-companion. Pairing is stored separately under paired-companion. Use Disconnect companion in FluxBench to revoke that connection before removing its saved configuration. The original owner’s legacy companion uses a different folder.

Connect Anthropic Claude

  1. Create your API key in Claude Console and arrange API billing there.
  2. Open AI & connections in FluxBench and choose Anthropic Claude.
  3. Paste your key in the API key field. If needed, enter a model ID available to your API account in Model override, then select Save settings.

API usage is billed separately by Anthropic. FluxBench does not currently connect to a local Claude Code subscription. Anthropic’s integration guidance directs third-party products to API authentication unless separately approved. A Claude chat login is not an API key.

Supported inputs and privacy

The paired connection accepts text and inline PNG/JPEG/WebP images within a 2 MB request. PDF text extraction is optional; see SETUP.md in the download. External attachment URLs, local file paths and tool execution are unavailable. Codex sign-in stays on your computer; the website holds a hash of a separate companion credential. Pending requests and complete replies are encrypted in the application database.

Requests expire after 230 seconds. Once claimed, a request is never delivered again. Expired stored payloads are removed on the account’s next companion interaction; deletion is not a scheduled retention service. Subscription costs and limits are distinct from per-request API billing. Additional-user and cross-platform pilots remain pending.

AI setup and companion downloads | FluxBench