FAQ & Tips

Common questions about Dices — pricing, keys, agents, privacy, parallelism, mobile, and your own servers — plus practical tips & tricks.

Answers to the questions people ask most about Dices — what it costs, what it needs, where your code goes, and how to get the most out of it. The second half is a set of tips & tricks worth skimming even if you think you know the app.

FAQ

Is Dices free?

Yes — Dices is free during alpha.

Do I need an API key?

To run Claude, yes — you need your own Anthropic API key (create one at console.anthropic.com; it’s billed to you and stored encrypted). Other agents authenticate in their own official CLI: you install the vendor CLI and sign in there, and Dices connects over ACP (Gemini also needs a GEMINI_API_KEY). So the answer is: you need credentials for whichever agent you choose to run.

Note: Dices never handles your provider credentials beyond your own API key — Claude uses your Anthropic API key; other agents authenticate in their own official CLI.

Can I use my Claude Pro/Max subscription?

No. Anthropic’s Consumer Terms don’t allow using a Claude Pro/Max (or Claude Code) subscription login inside third-party apps. Bring an Anthropic API key instead — create one at console.anthropic.com and you only pay for what you use.

Which agents and models can I use?

Dices is a bring-your-own-agent cockpit. It supports Claude, Codex, Gemini, Cursor, Hermes, and Pi, and many models across them. You can pick the right agent and model per task — and even per Folder (see parallel runs).

Does my code leave my machine?

No. The agents run locally, voice transcription runs locally (on-device Whisper by default), and the codebase index runs locally (local embeddings). The only thing that goes out is the model API calls you authorize — your code travels only as the context you send to the model. See Privacy & Security for the full model.

Can I run agents in parallel?

Yes. Dices is built for it — a grid view, a dockable workspace, and multi-model per folder let you run several agents at once. Different chats in the same Folder can even run different models, so you can fan a problem out across agents simultaneously.

Can I use my phone?

Yes. Dices ships as a PWA you can reach over Tailscale, so you can check in on your agents, respond to “needs input” prompts, and kick off work from your phone.

Does it work with my own servers?

Yes. Dices can drive your own machines over SSH using tmux, so your agents work where your code and infrastructure already live.

What makes it different from Cursor or an IDE?

Dices is a cockpit for directing agents, not a code editor. The workflow is brainstorm → build: you think through the problem, then hand it to agents that do the work. It runs on your infrastructure and the setup is fully yours. It is not a code editor — though it does include a light file viewer for when you want to glance at what changed.

Where are my API keys and passwords stored?

Secrets — API keys and SSH passwords — are encrypted at rest via the OS keychain and are never exposed back to the UI (the interface only knows whether a secret is “saved”). See Privacy & Security.

Is the remote server exposed to my network?

By default the remote server binds to Tailscale (encrypted) and loopback only — not the open LAN — and it’s token-gated. Opening it to your LAN is an explicit opt-in.

Tips & tricks

A grab-bag of habits that make Dices faster and safer to work with.

1. Use Plan mode first

Before letting an agent change anything, run it in Plan mode so it proposes an approach you can review. Approve the plan, then let it build — far fewer surprises than diving straight into edits.

2. Hand a subtask to a cheaper model

Not every step needs your most capable model. Spin off a chat on a cheaper model for the mechanical parts (boilerplate, renames, simple tests) and save the expensive model for the hard thinking. Multi-model per folder makes this trivial.

3. Pin per-project rules

Set per-project rules so every chat in a Folder inherits your conventions — code style, frameworks, “always run the linter” — without you repeating yourself in each prompt.

4. Install a skill

Add a skill to give your agents a reusable, packaged capability. Skills are previewed with security flags before install (see Privacy & Security), so you can vet them first.

5. Use the terminal’s quick-ask

In the terminal, use ⌘L / Ctrl L to pull what’s on screen into the agent — great for “what does this error mean?” or “fix this failing command” without retyping anything.

6. Set hooks to auto-run tests

Configure hooks so tests (or your linter, or a build) run automatically at the right moments. Let the machine catch regressions instead of remembering to run them yourself.

7. Brainstorm, then Build

Lean into the brainstorm → build flow: talk the idea through with an agent first, then hand the whole thread to a coding chat. The plan you reasoned out becomes the spec the builder follows. See Brainstorm → Build.

Make Dices yours with theming, then share your setup so a teammate (or future you) can reproduce the same environment. A consistent cockpit is a faster cockpit.

Tip: The single biggest speed-up is the command palette (⌘K / Ctrl K) — search chats and message history, run runbooks, and start from an issue without touching the mouse. See Commands & the Palette.

Next steps

Privacy & Security — the honest, detailed version of “does my code leave my machine?”
Commands & the Palette — the power-ups behind several of the tips above.
Integrations — start work from Linear and GitHub issues.