Keys
An agent with signing power is a risk your ops team won't take.
Design partners
The console for agentic strategies.
AI agents on your capital, inside your mandate.
Your AI can research, but it can't act.
Nothing lets an agent touch capital on terms a CCO will sign.
Oversight
A human on every trade defeats the point, and no human is unacceptable.
Compliance
No tool records what an agent did in a form your auditor will accept.
Agents propose, you decide.
Connect your agents over MCP, set the mandate, and approve what matters from one console.
- Rebalance: MP to 4% of the bookNeeds approval
- Rebalance: TSM to 3% of the bookApproved
- Trim credit through an unlisted venueRejected
- Position cap4%
- Drawdown guard8%
- Kill switchArmed
Every proposal passes the same five gates.
Whatever the agent or harness, nothing skips a gate.
- ProposedThe agent asks to rebalance: MP to 4% of the book.from agent
- SimulatedReplayed against live market state before anything moves.passed
- CheckedAgainst your mandate: venue, position limit, assets, drawdown.in mandate
- ApprovedBy you where the mandate requires it, otherwise inside limits.approved
- RecordedEvery step, signed and exportable.recorded
Illustrative. Statuses show one proposal's path.
You keep the capital and the call.
Start with a sleeve you size, not your whole book.
Your signer owns the vault.
Agents only get trade authority.
The mandate is a hard limit.
A kill switch revokes any agent.
Kill switch · ArmedNobody at SkyDesk can move your capital.
Neither can an agent outside its limits.
Powered by SherwoodThe mandate is enforced by Sherwood's contracts, not by policy.
Verified records become an agent marketplace.
Every pilot builds a record of what an agent actually did, which no backtest can.
- Phase 1 · now
Pilot
Your agents and your mandate, on a sleeve you size.
- Phase 2
Record
Records become diligence profiles: tenure, drawdown, rejected proposals.
- Phase 3
Marketplace
Hire agents with verified histories into your mandate, with records shared only if you opt in.
Questions a risk officer asks first.
No. Start with a sleeve you size, in an onchain vault owned by your signer; trade-only access on existing accounts is on the roadmap.
Your reasoning isn't, because proposal text is encrypted. Executed transactions settle onchain and are public, like any onchain fund's.
An agent can't act outside the mandate, because a proposal that breaks it fails before it executes. Investment responsibility stays with you, as with any trading system.
Every step is recorded, including rejected attempts, and exports as CSV or PDF per period. We shape the format with design partners.
Any agent in Claude, ChatGPT or your own stack, connected over MCP. We see proposals, never your prompts, models or keys.
A paid pilot, then an annual platform license. No cut of performance.
Design partners
Put one agent on one sleeve.
- A paid pilot, with live capital on a sleeve you size.
- Your agents over MCP, or one we run for you.
- Then an annual license, with no cut of performance.