Command

tent models

Lists every model of the project with what it can do, and with test tries each fixed call of a model against its real server. Nothing calls a provider by itself: test is run by hand.

Usage

Text
tent models
tent models test [name] [--data <folder>]

Options

OptionMeans
testSends one small request for each fixed call that a model has - reply, fill, embed, hear, speak - to its server, with the key of its secret, and says which the server answered
[name]Of test: the model to try. Left out, every model of the app is tried
--data <folder>Of test: the data folder of the engine to ask. Left out, it is .tent/ of the project, and with no engine running there one is started for the length of the command, on the development data

Example

Text
tent models test writer

Output

With --json, tent models prints each model: its name, its provider and its name there, its abilities and the most tokens one request may hold, as the definition states them or the list of its provider knows them, its voices, whether it is a function, its limits and its fall-backs:

JSON
{ "ok": true, "value": [{ "name": "writer", "provider": "anthropic", "model": "claude-sonnet-5-5", "abilities": ["reply", "tools", "typedAnswers", "images", "reasoning"], "takes": 200000, "function": false, "limits": { "perDay": { "tokensOut": 2000000 } }, "fallback": ["backup"] }], "findings": [] }

tent models test prints, for each model it tried, each call with whether its server answered it, and the code and the words where it did not, and for a model with the ability reasoning whether its reasoning arrived:

JSON
{ "ok": true, "value": [{ "name": "writer", "calls": { "reply": { "ok": true }, "fill": { "ok": false, "code": "T7110", "message": "The model writer did not accept the request: ..." } }, "reasoning": true }], "findings": [] }

Every request is counted and logged as any model call is, and a limit of use holds it too.

Errors

code.T0001 where test names no model of the app.

See also

word.model, adapter.Provider, code.T7001