Cornerstones uses a five-layer model so agents can understand what a response is for before they act. The layers are product boundaries, not marketing categories: each one tells the consumer whether it is reading market facts, market structure, interpretation, evidence, or a decision-ready context packet.
Start with Help
Use the public client first when you are onboarding an agent:
python -m pip install cornerstones-client
cornerstones-client --help
cornerstones-client guide
Then use this model to decide which class of surface the agent should call.
Market truth
Read-side market surfaces such as quotes, bars, charts, profiles, sessions, indicators, and public crypto market reads.
Use this layer when the agent needs to know what happened or what is currently observable.
Market structure
Higher-value structure such as orderflow, promoted liquidity metrics, stock depth, stock tick, imbalance snapshots, and explicit live/historical semantics.
Use this layer when the agent needs to understand pressure, liquidity, or internal market movement rather than just the last price.
Intelligence contract
Macro, cross-asset, sentiment, geopolitics, Polymarket discovery, and other interpretation layers packaged for machine consumption.
Use this layer when the agent needs broader regime or catalyst context around the market facts.
Evidence and events
Evidence feeds, canonical event browse/export flows, alerts subscriptions, delivery receipts, replay, and recovery workflows.
Use this layer when the agent needs to explain why a move matters, export events, subscribe to relevant updates, or attach evidence to a decision path.
Agent context
Decision-ready context packages meant to plug into downstream agents, research loops, and execution systems. Current context surfaces include FX, Gold, Stocks, Cross-Asset, Polymarket, and Geopolitics paths where available.
Use this layer when the agent needs a compact, provenance-aware packet rather than stitching lower-level reads itself.
Consumption rules
- Always inspect
provenance,degraded,fallback, andnot_implemented. - Treat bounded surfaces such as
stocks universe,stocks screener, andstocks optionabilityas workflow inputs, not unlimited exports. - Treat Max-only and premium-depth surfaces as controlled slices, not default anonymous access.
- Use
guideandchangelogfor current capability discovery before assuming a surface exists.
Website note
This secondary model helps orientation. It does not claim a final public pricing table or unlimited anonymous access to the managed runtime.