AI / MCP
Give your agents the same contracts as your best engineers.
Model Context Protocol is the spine: Chronix exposes tools for setup, diagnostics, simulation, and guarded change management. Your models never need ad-hoc SSH recipes — they inherit policy, auditing, and rollback paths.
Primitives
Operators first. Autonomy second.
Chronix ships tool surfaces with explicit capabilities: what can be called, what must be human-approved, and what automatically halts trading when probes fail.
- Tool contracts: versioned MCP endpoints that map to real provisioning and risk systems — not HTML wrappers.
- Private LLMs: deploy inside VPC, colo, or sovereign regions with identical interfaces to cloud inference.
- GUI composer (preview): agents turn natural-language desk intents into concrete pages—see the section below. Not production-ready yet.
Preview
MCP GUI composer — layouts from intent
The original MCP plan includes tools that fabricate customized operator GUIs: you describe the job, Chronix emits a page with the right widgets, wires it into the backend, and registers it in the pages collection consumed by the WASM/C++ desk shell. This path is work in progress and not ready for production workloads—reach out if you want to co-design it.
- Example spread desk: “I am pricing a basis spread—give me three depth-of-market widgets, one per leg, plus a synthetic spread DOM, plus a spread P&L chart.” The composer allocates layout regions, binds feeds, and publishes the page to your tenant.
- Backend + WASM: generated views land in a backend pages store and appear in the client’s page collection so the native shell (WASM with C++ hot paths) can host them without one-off HTML hacks.
- Guardrails: the same policy and audit envelopes that protect MCP tool calls apply to GUI materialization—human approval hooks stay first-class.
// Operator prompt → MCP tools → page record + desk registration
request({
workflow: "spread_monitoring",
widgets: [
{ type: "dom", bind: "leg:A", title: "Leg A DOM" },
{ type: "dom", bind: "leg:B", title: "Leg B DOM" },
{ type: "dom", bind: "leg:C", title: "Leg C DOM" },
{ type: "dom", bind: "spread:synthetic", title: "Spread DOM" },
{ type: "chart", bind: "spread:pnl", title: "Spread P&L" },
],
persist: "pages_collection",
runtime: "wasm_cpp_shell",
}); Example
What an agent sees (and what it never should).
{
"server": "chronix/on-prem",
"tools": [
{ "name": "chronix.cluster.health", "risk": "read" },
{ "name": "chronix.strategy.promote", "risk": "write", "require_human": true },
{ "name": "chronix.marketdata.pause", "risk": "halt-capable" }
],
"policy": {
"halt_on_tool_error": true,
"secrets": "vault:chronix/prod/east"
}
} Engagement
Bring your model team and your trading ops leads.
We’ll trace a real change — from idea to sim to production — entirely through MCP-safe workflows.