Co-pilot vs autonomous
Two ways a desk can run: co-pilot proposes and waits for your approval on every trade; autonomous trades on its own inside the limits. Both face the same risk engine.
Updated 2026-07-17
Every desk runs in one of two modes, switchable at any time from the desk page:
| Co-pilot | Autonomous | |
|---|---|---|
| Who pulls the trigger | You: every trade waits for explicit approval | The AI: trades on its own |
| Risk engine | Checks at proposal time AND re-checks at your approval | Checks every entry and manages every exit |
| Speed | As fast as you approve | Acts the moment a setup clears the bar |
| Best for | Learning, trust-building, high-stakes accounts | The full product: a desk that works while you sleep |
How co-pilot approval works
Proposals wait in a queue on your dashboard and desk page, each with the full thesis, size, stop, and the engine's verdict already attached. When you approve, the engine re-validates at that moment against your live account (prices moved, margin changed, maybe another desk traded), so a stale proposal can never fill on yesterday's numbers. Vetoed-at-approval trades tell you which rule fired.
What autonomous does not mean
- It does not mean unlimited: the same 25 rules gate every entry, and exits (stops, take-profits, trailing) are engine-managed, deterministic policy, not AI whim.
- It does not mean unattended-forever: auto-pause and daily-loss locks stop the desk on bad streaks, and every action lands in the decisions feed and your daily digest.
- It does not mean irreversible: pause, per-position close, and the kill-switch always outrank the AI. See Close, pause, and the kill-switch.
A good pattern: start co-pilot on a practice desk, approve a dozen trades to calibrate your trust, then flip autonomous, first on testnet, then live.