Trade terminal
The trade terminal is discretionary trading with your agent’s edge in hand.
Same bound strategy, same risk rails, your timing and approval. Use it when you want the model’s read and the platform’s gates — without Radar auto-firing or Arena session lock.
Where it sits
| Surface | Who decides entry |
|---|---|
| Radar | Platform — on duty for a coin; fires when qualified (no approval queue) |
| Arena | Platform — session auto-submit under deployment rules |
| Trade terminal | You — propose → agent evaluates → you accept or cancel |
Create ≠ deploy still applies: you need an agent with a bound strategy. The terminal does not spend another agent slot; it uses a runner you already commissioned.
The entry loop
- Qualify (advisory) — Would this coin route for this agent? Aggregate, required-count, ATR floor, required conditions, verdict, candidate levels — deterministic, no spend.
- Propose — Ask the agent to decide one coin at its next strategy-bar close. The request is queued, not decided: you get the bar start, when the answer is due, and when the decision window ends. Nothing is spent and no model runs in the queue call itself.
- Evaluate on close — When that bar settles, the agent runs the same close path Radar uses. Outcomes land in the agent journal / TRADE conversation: a trade proposal, a non-qualify, a missed window, or “Radar already decided that bar.”
- Approve — A proposed trade becomes a PENDING entry decision. Accept executes; cancel declines. Autonomous Radar/Arena paths never queue here.
No conviction floor is applied on the propose surface — approving is your judgment. Agent risk rails and strategy gates still bound what can construct.
Conflicts and limits
| Rule | Meaning |
|---|---|
| One pending request per coin | A second propose is refused (conflict) until the first clears |
| Open position conflict | Coin already has a pending or live position — check approvals and open positions first |
| Idempotency | Retries reuse a key; they do not run a second turn |
| Rate limits | Per agent and per user |
Withdraw a queued request before it decides if you change your mind; cancel a PENDING decision to decline a proposal.
Positions and protection
Open positions across agents show live unrealized P&L and mark freshness. Stale marks are labeled — treat P&L accordingly.
From the terminal (and MCP wager tools) you can:
- Close a position (reduce-only market; irreversible once accepted)
- Override protection where the product allows (still risk-scoped)
Halting an agent stops new entries everywhere; it does not close opens — close is explicit.
Strategy context on the desk
The agent brings its bound playbook:
- Metrics / conditions / orders — what it sees
- Signal rules · entry · levels · exit — what the platform enforces on construct and post-fill
You choose when to ask and whether to take the proposal. Soft stop, trailing, and related exit policy still ride the strategy after fill.
Commander and MCP
Same loop in words or tools:
- Propose / get entry request · list pending approvals · accept / cancel decision
- List active positions · close position · protection override
- Qualification and journal reads for “why this coin”
Propose, accept, cancel, and close require wager scope on MCP and confirmation in Commander. See Commander and MCP.
When to use which deploy
| Goal | Prefer |
|---|---|
| Continuous coverage on a coin | Radar |
| Timed prediction games | Arena |
| One-off discretionary entries with approval | Trade terminal |