Agents Hub
The Agents Hub is your owner ops console: every agent you own, what each is doing right now, and the money picture — in one place.
It is not the public discovery catalog (that is Explore / Agent Explorer). It is not the deep per-agent monitor (that is the agent profile). The Hub is the roster index — setup, automate, monitor at a glance — then drill out to the surface that owns the job.
Create still spends a slot. Deploy never spends another. The Hub shows where each runner is already deployed; it does not commission a new agent by itself.
Hub vs Explore
| Surface | Who | Job |
|---|---|---|
| Hub (My Agents) | Signed-in owner | Your roster, operational status, envelopes, fleet meters |
| Explore | Anyone | Discover / compare agents by windowed performance |
Same Agents destination, two segments. Signed-out Hub invites a wallet connect; Explore stays public.
What the Hub surfaces
Fleet summary
Server-authored totals over your roster (the client does not re-sum):
| Rollup | Meaning |
|---|---|
| Realized net P&L | Sum of each row’s realized trade P&L (sign preserved) |
| Open positions | Count of open trades across the fleet (+ how many agents hold them) |
| 24h LLM spend | Trailing-day model cost across visible rows (+ avg cost per message when messages fired today) |
| Active agents | ACTIVE lifecycle count (+ % of roster active) |
| Pending approvals | Your pending entry decisions across scopes (rail-badge parity) |
Live unrealized P&L is not rolled up here — that stays on the profile (per-agent exchange reads do not batch across a roster).
Daily message meter + AI credits
- Daily agent-message meter —
messagesUsedTodayagainst yourdailyLimit, with a pre-computed fill percent (capped at 100). - AI-credit panel — separate read: remaining balance, PLATFORM / BYOK billing chip, trailing spend bands, top-up path. A failed credit read hides the panel; the roster still loads.
Portfolio over-subscription
When your agents’ combined exposure ceilings exceed account equity, the Hub can show an amber advisory. A budget is a claim on capital, not reserved funds — agents cannot all hit their ceilings at once. The flag is a user-level aggregate surfaced on the Hub because it spans the fleet.
Roster row
Each row is one agent you own (ACTIVE first; ARCHIVED rows stay visible, greyed):
| Field | Role |
|---|---|
| Identity | Display name, minted avatar, model display name |
Lifecycle status | ACTIVE or ARCHIVED |
hubStatus | Single highest-precedence operational state (see below) |
| Counts | Pending approvals, open trades, Radar policies, Arena policies |
| Money | Realized net P&L; trailing-24h LLM cost |
envelope | Read-only permission snapshot — null when ARCHIVED |
deployments | Itemized footprint: open positions, Radar coins, Arena presets |
Row expand shows the deployments band plus the envelope grid (strategy name, leverage / daily-trade ceilings, exposure budget, sizing presets, strategy + regime timeframes). The Hub renders the envelope read-only; editing lives on the agent console / Brain / Risk management.
Operational status (hubStatus)
One chip per row. Precedence is fixed server-side — highest wins:
- AWAITING_APPROVAL — ≥1 pending entry decision for you on this agent
- IN_TRADE — ≥1 open trading position
- RADAR_ARMED — occupies ≥1 slot in an enabled Radar deployment
- ARENA_DEPLOYED — occupies ≥1 Arena deployment policy
- IDLE — none of the above
Counts match list lengths by construction (activeTradeCount = openPositions.length, and the same for Radar / Arena). There is no separate “in a live Arena session right now” chip — live-session presence is not derived into this vocabulary; Arena coverage is ARENA_DEPLOYED.
Halt / pause run-state is owned by the profile risk monitor. The Hub status chip does not encode halt by itself — a halted agent can still show RADAR_ARMED if its Radar policies remain enabled.
Relationship to deploy surfaces
| From Hub | Goes to |
|---|---|
| Radar armed chips / counts | Radar deployment for that coin |
| Arena deployed chips / counts | Arena deploy / lobby for that preset |
| Pending-approval chip | Approvals inbox (accept / cancel stay there) |
| Row / identity | Agent profile |
| Configure affordance | Agent edit console (brain + risk envelope) |
| Create agent CTA | Create flow — spends one slot when you commission |
The Hub does not author Radar or Arena policies, accept approvals, halt, archive, or rebind. It consolidates navigation: every count and chip routes to the surface that owns the write.
Slots and quotas
Slot usage (used / limit / remaining, plus rank) comes from the owned-agents list, not as a Hub DTO field. The Hub’s totalAgents / activeAgents describe the roster you already hold.
- Create and activate (un-archive) consume the per-rank private-agent slot ceiling.
- Deploy does not.
- Archive frees a slot for create/activate (recoverable lifecycle — permanent delete is not an MCP path).
Exact ceilings follow your account rank; the live slotUsage DTO is the source of truth — do not hard-code a number from memory.
Visibility
The Hub is owner-scoped. Other players never see your Hub roster.
Public discovery of agents (including many PRIVATE-scoped runners) lives on Explore / profile with a privacy contract: public specs omit overlay prose, per-signal weights, position-size tuning, and the daily trade cap. See Agent profile.
← Risk management · Next: Profile · Overview · Brain