Skip to content

solvi.cli

The solvi command line (python -m solvi ...): one module per command group.

The solvi command (also python -m solvi ...).

solvi test PATH...        decision regression tests from cases.json files (solvi.testing)
solvi honesty SET.json    honesty numbers of a labelled set, gated against a baseline (solvi.testing.honesty)
solvi mutants PATH...     small faults made in the catalog: which ones the cases do not catch (solvi.testing.mutants)
solvi verify | replay | diff over a TraceStorage:

solvi verify decisions.db [--anchor COUNT:HASH] [--signature SIG.json] [--sign SIG.json]
solvi replay decisions.db --system myapp.decisions:system
solvi diff   decisions.db --system myapp.decisions_v2:build_system [--question Q] [--since ISO] [--limit N] [--json]
solvi serve  myapp.decisions:system [--store decisions.db] [--decider ID] [--port 8000] [--mcp]   (solvi.serve)
solvi serve  --guard catalog.py:guard --upstream "MCP SERVER COMMAND" [--store calls.db]         (solvi.serve.run_proxy)
solvi check  myapp.decisions:system [--strict] [--json]                                          (solvi.check)
solvi report decisions.db [--since ISO] [--until ISO] [--question Q] [--id ID] [--html out.html] [--md out.md] [--json]
                                                                                                 (solvi.core.store.report)
solvi report decisions.db --overview [--since ISO] [--until ISO] [--question Q] [--drift-window N | --no-drift] [--json]
                                                                                                 (solvi.core.store.sysreport)
solvi init [DIR] [--template support|refunds|minimal] [--with-model] [--force]                  (solvi.cli._scaffold)
solvi ask myapp.decisions:system (STATE.json | - | --state '{...}' | --text "...") [--question Q] [--decider MODEL]
          [--audit] [--report md|html] [--lang ru] [--store decisions.db] [--json]
solvi calibrate myapp.decisions:system PART labels.csv --risk 0.1 [--groups a,b] [--method crc|ltt] [--out F]
                                                                                                 (solvi.cli._calibrate)
solvi models [list | pull ID | check MODEL --examples labels.jsonl --task Q]                       (solvi.cli._models)
solvi hook [install | uninstall | pre-edit --rules rules.toml | pick-skill --skills-dir DIR]        (solvi.experimental.hooks)
solvi migrate PATH [--check]     rewrite the 0.9 import paths of your code to the 1.0 ones          (solvi.cli._migrate)

--system names a System: "package.module:attribute" or "path/to/file.py:attribute", where the attribute is a System or a function without arguments that returns one (solvi serve and solvi check take it as their first argument; check also takes a Catalog). solvi report prints a Markdown report of the stored decisions (or of one with --id), or writes it as a self-contained HTML page (--html) / a Markdown file (--md); --system is optional there (typed values, the replay of one decision). solvi ask asks once (a state, or a text through ask_text) and prints the answers, the audit or a report; a module-level prepare(state) next to the System runs first, as in solvi test. Exit status: 0 — verified / everything replays / nothing changes / no catalog errors / report written / scaffold written / every asked question answered / something answered alone after calibrating; 1 — problems / mismatches / changes / catalog errors / a file exists / a question abstained / the calibration escalates everything / a model that does not load; 2 — usage errors.