Documentation
¶
Overview ¶
Package cli implements every mpqt command against injected dependencies. It is llm-free; cmd/mpqt (the nested module) supplies the constructors (App.NewClient, App.NewCounter) that reach a real inference.Client and pricing.Counter. This package never imports github.com/looprig/llm and never constructs a client or counter itself -- it only ever calls the functions App carries.
Index ¶
Constants ¶
const ( ExitOK = 0 ExitCommandFailure = 1 ExitUsage = 2 ExitGateFailed = 3 ExitPricing = 4 )
Process exit codes, per Main's own doc comment: 0 ok; 1 command failure; 2 usage; 3 disposition/comparison gate failed; 4 cost ceiling or pricing-completeness failure.
Variables ¶
This section is empty.
Functions ¶
Types ¶
type App ¶
type App struct {
Registry *packfile.Registry
NewClient func(model.Model) (inference.Client, error) // nil => gen/run/judge unavailable
NewCounter func(model.Model) (pricing.Counter, error) // nil => heuristic preflight
LookupEnv func(string) (string, bool) // for key presence checks (never values in output)
Stdout, Stderr io.Writer
Now func() time.Time
// RateLimit configures client-side rate limiting (RPM pacing, concurrency
// cap, 429/5xx retry-with-backoff) applied to every live client this App
// hands out. The zero value disables it (a pure passthrough). The paid
// commands set it from their --max-rpm/--max-retries/--max-concurrent-
// requests flags before resolving any client, so both the target and the
// judge client route through the same limiter.
RateLimit ratelimit.Config
}
App wires the environment-specific pieces. Every field has a working zero-cost default except the client constructors, which nil out LLM commands with a clear error: App{} is enough to run init/validate/schema/ evaluators/compare (see withDefaults), but gen/run need NewClient (and, for a real token estimate rather than the heuristic fallback, NewCounter).
type LLMConfig ¶
type LLMConfig struct {
LLM struct {
Provider string `yaml:"provider"`
Model string `yaml:"model"`
APIFormat string `yaml:"api-format"`
BaseURL string `yaml:"base-url"`
} `yaml:"llm"`
}
LLMConfig is the generator/judge configuration file shape: non-secret provider/model identity only. The API key comes only from the provider's own environment variable (never this file, never a CLI flag) -- see checkKeyPresence.
Source Files