Commands

Cents exposes 22 top-level command groups, each grouping related operations on theses, research, watchlists, alerts, positions, broker state, the autonomous factory loop, pre-registered experiments, calibration, and the LLM eval harness.

The pages below are auto-generated from the CLI’s --help output, so they stay in sync with the installed cents binary. Each page lists subcommands, options, types, defaults, and a usage example.

Generator status

Per-command pages are populated by scripts/generate_docs.py (Phase 2 of the launch plan). Until that generator runs, the cards below link to pages that may not yet exist — start with the Quickstart for runnable examples.

Command groups

thesis Create, list, update, and close investment theses.

research Run research agents on a symbol; export reports to JSON or HTML.

evidence List, inspect, and curate evidence attached to theses.

watch Manage the watchlist (add, list, remove, set thresholds).

scan Scan the watchlist and emit alerts when conviction shifts.

alert View and manage alerts produced by recent scans.

position Open, close, and inspect positions linked to a thesis.

outcome Record outcomes against a thesis and review accuracy.

recommend Emit model signals (bullish/bearish/neutral) for open theses. Not investment advice.

portfolio Manage portfolios as separate database files.

broker Alpaca broker integration — positions, orders, sync.

backtest Run and analyse historical agent backtests.

status Show current configuration and database status.

factory Run and inspect the autonomous open/close loop over a symbol universe.

universe Define and manage the symbol universes the factory walks.

screener Run discovery screeners (value, growth, momentum, mean reversion, insider cluster).

cohort Inspect directional vs neutral cohort spread P&L.

event Refresh and inspect policy events used for premise invalidation.

experiment Pre-register hypotheses with a frozen factory.toml so analytics stay falsifiable.

calibration Fit and inspect the logistic-regression conviction calibration model.

eval Run the LLM eval harness against golden fixture sets.

usage Summarise persisted LLM usage and cost by agent / model / day.

How regeneration works

The generator walks the Click command tree (cents.cli.cli.commands) and emits one MDX file per group into this directory. Hand-written prose lives in sibling *.intro.mdx files that get prepended at generation time, so regenerating after a CLI change never clobbers the narrative.