Router
Classifies intent and routes to the right tools.
Compass turns institutional research, data and investment frameworks into governed, testable intelligence for investment teams.
Compass follows a 5-step graph to deliver accurate, traceable and actionable responses, with human oversight and enterprise controls at every step.
Classifies intent and routes to the right tools.
Selects and orders the best tools for the task.
Runs analytical tools and collects results.
Cites every number with source provenance.
Streams a cited, annotated reply to the user.
Six specialized tools work together to turn complex questions into actionable investment intelligence.
Price & yield forecasts for equities and bonds.
Cross-sectional ranks across stocks and credits.
7-arm comparison: HRP, CVaR, BL-Alpha and more.
Multi-axis regime: rates, curve, credit, VIX, styles.
Internal research, textbooks & filings, cited.
Holdings review: score, swap & forgone alpha.
Illustrative thread — interface mock-up, not a performance claim. No figures shown are back-test or live results.
PM asks for the regime and its tilt implications; Compass cites market_data, regime and risk scenario tools.
What is the current market regime and what are the tilt implications?
the current regime is “Risk On” with a positive equity tilt, Key drivers include rising growth, easing inflation and stable credit spreads.
A proactive forecast pack surfaces conflicts between ranking and forecast models; the analyat drills into drivers.
Show me the 1-month price forecast for AAPL and the key drivers.
AAPL has a +4.85% forecast return. However, the ranking model is less bulish due to sector rotation. Key drivers earning momentm, buybacks and improving demand.
The PM edits an optimized book in natural language; the mandate gate re-checks and the graph resumes from the checkpoint.
Increase US tech exposure by 5% and keep within the mandate.
Updated portfolio weight applied. Mandate and restricted list check passed. Optimization complete. Resuming from checkpoint.
Flags swap candidates, cites the published methodology and queues proposals for approval.
Run the True North Review for this quarter.
12 swap candidates identified. Top 3% TSM, BRKB, and CRM. See methodology and proposed changes in the report.
From the core tools every thread uses, to modules built for a specific desk’s decisions.
Any of these — and any of your own REST systems — becomes a Compass tool from a JSON spec at runtime: wired by configuration, not custom integration.
Problem and outcome: Stop stitching together research, models and notes. Ask one question and get cited answers, with every number computed by a deterministic, point-in-time tool and no language model in the scoring path.
Description: Each module answers a specific desk’s decisions, with its own page, buyer and call to action.
TRUE NORTH — MODULE DETAIL
Scores every position, proposes like-for-like swaps inside the mandate, and reports the alpha left on the table.
TENOR — MODULE DETAIL
Which loans will be recalled, how long they will last, what is deliverable at settlement, and what rate to quote — each scored with the reasons behind it, from your own loan, inventory and locate history, nightly or on demand, entirely inside your VPC.
LODESTONE — MODULE DETAIL
Classical forecasts break when markets cluster. Lodestone forecasts volatility, correlation, network shape and co-movement scenarios for about 500 liquid US stocks at 1-week, 1-month and 1-quarter horizons, so risk, hedges and stress tests reflect how stocks move together.
Compass runs on frontier models (Claude / GPT-class) under Zero Data Retention — the model is a component, not the product. Model-agnostic by design.
Local deployment; remote models under Zero Data Retention; per-desk entitlements and information barriers honoured.
Any internal REST system becomes a tool from a JSON specat runtime; no MCP servers to build, no new outbound network connections.
Mandate & restricted-list gate as a graph node; every number grounded; human approval with checkpoints; append-only audit.
Full trace behind every answer; golden test sets, replay, drift detection; vendor upgrades become measured, reported changes.
Versioned, registrable (MRM).
ENGAGEMENT MODEL
Structured, measurable and designed for long-term value.
A pragmatic path to value, with the pilot as the center
Quantitative intelligence uncovers the insight. PredictNow.ai helps transform that signal into better decisions.