Skip to main content
The Agentic OS for Capital Markets

The agent harness purpose-built for investment work

Dominion is the agentic operating system for capital markets — a steerable agent harness that orchestrates a fleet of specialist sub-agents across an integrated tool registry of spreadsheets, market data, filings, and macro feeds. Plan-mode reviews. Build-mode execution. Headless monitoring. Citation-grade reproducibility on every run.

The Challenge

  • General-purpose chat assistants hallucinate figures and have no harness to verify the math against the source model
  • Spreadsheet errors — broken refs, circulars, hardcoded plugs — slip into IC decks because no agent owns verification
  • Market data, filings, FX, and macro context live in five different tools with five different logins and no shared memory
  • Background monitoring lives in someone's spreadsheet of email rules; threshold breaches surface a day late
  • Reproducibility and audit trails are afterthoughts — making AI-assisted analysis non-starters for IC and compliance

Our Solution

  • An agent harness purpose-built for finance: orchestration, memory, tool registry, and permissioning baked in
  • A steerable sub-agent fleet — Analyst orchestrates, Auditor verifies, Quant computes, Sentinel watches, Scribe writes
  • A typed tool registry across spreadsheets, market data, filings, FX, and macro — no copy-paste, no context loss
  • Headless, long-running Sentinel loops watch artifacts continuously and escalate by severity, not by volume
  • Every run is reproducible by construction — inputs, code, seed, citations, and audit log retained for the desk and compliance

An Agent Harness Built for the Desk

Orchestration, sub-agent specialization, a typed tool registry, plan-mode and build-mode loops, headless background runs, permissioned safety surfaces — engineered for the way investment work actually happens.

A Steerable Sub-Agent Fleet

Dominion runs an orchestrated fleet of specialist sub-agents — the seats on a modern investment desk, exposed as composable building blocks. The Analyst is the orchestrator: it parses intent, plans the run, and delegates to specialists through a permissioned tool registry. Every hand-off, every tool call, every artifact is logged for review and re-run.

Key Benefits

  • Analyst — the orchestrator. Plans the run, decomposes the brief, routes work to specialists, and assembles the cited answer
  • Auditor — plan-mode verifier. Read-only recompute and reconciliation against the source model, with severity-rated findings
  • Quant — build-mode specialist. DCF, Monte Carlo, sensitivity, portfolio optimization, and backtesting in a sandboxed Python and R runtime
  • Sentinel — autonomous, headless. A long-running loop that watches positions, models, and feeds, paging the desk only when severity warrants
  • Scribe — output agent. Memos, IC decks, tearsheets, and updates rendered with citations bound to source cells, filings, and feeds

The Agentic Loop, Applied to Investing

Plan, execute, verify, monitor, ship — Dominion runs the same loop that defines modern agentic systems, retooled end-to-end for the realities of capital markets.

Step 1

Mount the Desk

Point Dominion at your workbooks, wire in the market, filings, FX, and macro connectors, and define the permission grid for each sub-agent. Runs local-first or in your VPC — your artifacts never leave the perimeter you set.

Step 2

Brief the Orchestrator

State the objective in plain English. The Analyst plans the run — decomposing the brief, selecting tools, and routing sub-tasks to specialists. You see the plan before it executes.

Step 3

Verify in Plan Mode

The Auditor recomputes the model independently in a sandboxed read-only scope. Discrepancies, broken refs, and convention breaks surface as severity-rated findings before any write happens.

Step 4

Execute in Build Mode

The Quant runs DCF, Monte Carlo, sensitivity, and goal-seek against the live workbook. Live quotes, fundamentals, options, filings, FX, and macro arrive through typed tool calls — every figure cited at the moment of retrieval.

Step 5

Hand Off to Headless

Sentinel takes the artifacts you've registered and runs as a long-running background loop on the agent server. Threshold breaches, model drift, and material filing changes escalate to the desk by severity, not by volume.

Step 6

Ship a Reproducible Run

Scribe compiles the IC memo, tearsheet, or update with citations bound to source artifacts. Export to PDF, Excel, or your firm's house template — and re-run the same artifact against fresh data with one command.

Intelligence layer

Move from scattered signals to capital-ready decisions.

Sig Analytica connects market research, relationship context, and deal workflows into one operating layer for private capital teams.