Every account gets its own AI agent. Yours knows your book.
Not a chatbot bolted onto a dashboard. A per-workspace agent with its own scoped credential, reading the same data your screens do, in the chat you already live in — Discord or Telegram.
Book: 4 open, net long $38.2k. NVDA 0DTE call is 8% from its premium stop; trail armed at TP1.
Providers: delta_desk is 5 signals from its 10% monthly breaker; harbor_stocks has 3 unmatched updates. Want the disposition queue?
| apex_signals | +4.1R | PF 1.9 · n=14 |
| delta_desk | +1.2R | PF 1.1 · n=22 |
| northstar_options | −6.3R | PF 0.5 · n=19 |
Slippage decomposed: gap 101 bps (the market was already through the level) + 15 bps execution. The order itself cost almost nothing — the cadence gap did. Want the shadow record for the near-level wake?
morning_market_checkdesk_session_opentrade_reviewprovider_rankingexposure_by_underlyingReady-made prompts ship with the MCP server. Example conversations are illustrative.
Pick the model. Hosted, or private to you.
The included agent runs on a frontier hosted model through a gateway that keeps your workspace credential server-side. Desk and Allocator plans can move the agent onto a private deployment of an open-weight model, so nothing about your book leaves infrastructure you approve.
Hosted frontier models
Routed through our gateway with per-workspace spend limits and full tracing. Switch models without changing anything about the agent's access.
Claude
GPTGPT
Gemini
Mistral
Private-deployment options
Open-weight models served from a tenancy you control — our managed cluster, your cloud account, or on-prem — fine-tuned on your fills, providers and playbooks.
DeepSeek
Kimi
MiniMax
GLMGLM
Qwen
Llama
Mistral
Ollama self-hosted
Hugging Face open weights
Model names and marks belong to their owners and indicate compatibility, not endorsement. Availability of a given model on a private deployment is confirmed when we scope the build.
Custom private-model agents, built on your book.
Beyond the included agent: we develop agents on a private model deployment trained and tuned on your workspace — your fills and outcomes, your providers, your risk settings, your research — with your own playbooks and reporting formats.
- Your data never trains a shared model.
- Same no-execution ceiling, unless you grant a narrower, audited write group.
- Scoped on a call; typical build 2–4 weeks; from $299 per agent per month plus a one-time build.
What a private agent typically owns
- A morning brief in your format, posted before the open
- Provider reviews on your criteria, with the move drafted for approval
- Scenario work on your book: "what if we halve options risk on CPI days?"
- Post-mortems on every stop-out, with the slippage decomposed
- Weekly allocator reports across every workspace you oversee
About the agents.
Can the agent place trades?
Not by default, and not by prompt — by permission. Every workspace agent runs under a server-side ceiling that cannot place an order, arm an entry monitor, change routing or touch credentials. If a desk wants an agent to act, a human grants a narrow, audited write group.
What does it read?
Everything the workspace can see through the documented API: positions, P&L, trade ideas and their event timelines, signal imports, provider performance, scanner and desk rows, market regime, macro calendar, news and research. It reaches exactly what its token reaches.
Where does it live?
In your Discord or Telegram today, with per-workspace roles. Bring your own agent (Claude, Hermes or your own) over MCP with the same ceiling.
How is a private-model agent different?
It runs on a private model deployment trained and tuned on your workspace's data and playbooks. Your data never trains a shared model. It is scoped on a call and billed as an add-on on Desk and Allocator.