Co-pilot
Co-pilot is VolatiCloud's assistant. You tell it what you want in plain words, and it works through the same operations the console uses: it reads your strategies, backtests and bots, builds strategies, runs backtests and hyperopt, and starts dry-run bots. This page covers what it can do, what it asks you before it acts, which plans include it, and what it costs.
Co-pilot is the same helper whether you chat with it or hand it a Mission. Chat is a conversation you are present for. A Mission is work it does in the background, within a budget you approve.
What Co-pilot can do in chat
- Answer questions about your account. It reads your strategies, backtests, bots and runners with your own permissions. Fields that hold secrets, such as exchange API keys, are removed before the model sees anything.
- Build a strategy. It drafts a strategy from your description and shows it on a card. The strategy is saved in your workspace only when you approve it.
- Run a backtest or a hyperopt. It shows a card with the run and its cost, and the run starts only when you approve it. Hyperopt is on Pro and Enterprise.
- Start a dry-run bot. One approval creates the bot and starts it. A dry-run bot trades with simulated money on live market data, with no exchange keys and no real funds. It can also start a dry-run bot you already have, after you approve.
- Explain what you are looking at. Pages for strategies, backtests, bots, exchanges, runners, hyperopt runs and simulations have an Ask Co-pilot entry that opens the conversation about that item.
Co-pilot never proposes a live bot. Every bot it offers to create is a dry-run bot.
What Co-pilot asks before it acts
Anything that spends credits, creates or changes a strategy, or changes how a bot runs goes on an approval card first. The card shows what will run, and nothing happens until you select approve. Cards also collect on the Approvals page: requests from Co-pilot and your Missions that wait for your decision.
| Co-pilot wants to | What happens |
|---|---|
| Read your data | Runs straight away, with your permissions |
| Create a strategy | Always a card with the strategy and its code; saved only after you approve |
| Save a new version of a strategy, or apply hyperopt parameters to it | Card showing the change; applied only after you approve |
| Run a backtest, a walk-forward or consistency-check backtest, a hyperopt, or a trade replay | Card with the cost in dollars; runs only after you approve |
| Create and start a dry-run bot | One card covers both; runs only after you approve |
| Start an existing dry-run bot | Card with what it costs to start; runs only after you approve |
| Change your organization's spending policy | Always a card, in every conversation |
| Rename a strategy, fork a marketplace listing into your workspace, rename a conversation, mark alerts as read | Runs straight away while the conversation has read only your organization's own content; a card once it has read content you did not write |
| Stop any bot, start a live bot, change exchange keys, billing, members or security settings, publish, or delete | Refused; you do it yourself in the console |
Content you did not write. Once a conversation has read something from outside your organization, such as a public strategy or a marketplace listing, Co-pilot no longer makes even the small changes in the table on its own: each one goes on a card, and the card says which outside content it read first. Reading a listing counts, so forking one is in practice always a card.
Only you can approve. The approval step does not call the AI model. It takes the operation from what the server stored when the card was made, and runs it as you, with your own permissions. Co-pilot has no way to approve its own card.
If your organization uses passphrase encryption, Co-pilot does not build, edit or iterate a strategy, or start a backtest, on your behalf. For those it answers and explains, and hands you off to the Strategy Studio to do it in your own browser session.
Plan access
| Plan | Co-pilot chat | Missions |
|---|---|---|
| Free | Limited: about 20 messages a day on a basic model (see below) | No |
| Starter | Yes, paid from credits | No |
| Pro | Yes, paid from credits | Yes |
| Enterprise | Yes, paid from credits | Yes |
Free Co-pilot chat is funded by VolatiCloud and never touches your credit balance. It has limits:
- It needs a verified email address.
- It uses the basic model only.
- It allows about 20 messages per person per day, which resets at midnight UTC.
- It answers questions about trading and VolatiCloud only.
- It can be paused for the rest of a month, when the platform's monthly budget for free chat is used up.
Pro and Enterprise also add larger AI models, and the option to use your organization's own model key (an OpenRouter key). See Plans and pricing for everything else each plan includes.
What it costs
On Starter, Pro and Enterprise, each Co-pilot message has a cost in dollars: what the AI provider charged to answer it, times 1.10, or times 1.40 on the largest (frontier) models. It is paid in credits from your organization's balance, and one credit is one US dollar (1 credit = $1). On Free, Co-pilot chat costs you nothing; VolatiCloud pays for it.
Three settings keep that spend in check:
- Co-pilot chat monthly cap. An organization admin can set a monthly cap for Co-pilot chat in the Spending policy. When it is reached, chat stops until the next month or until an admin raises the cap.
- Bot reserve. Part of your balance is reserved for running bots, about 72 hours of their running cost by default; an organization admin can change it in the Spending policy. Co-pilot chat, Missions and research never use it, so your running bots keep running.
- Your own model key. On Pro and Enterprise, when your organization sets its own model key, Co-pilot messages use that key and are not charged in credits.
Backtests and hyperopt runs that Co-pilot starts are charged like any other run, and their card shows the cost before you approve it.
Related
- Missions: background work within a budget you approve
- Credits: how the credit balance works
- Bot lifecycle: from dry-run to live
- MCP integration: connect your own AI agent instead