Method Chat

internal/llm/llm.go:286–290

Chat streams an assistant response on the returned channel, closing it when the stream ends. Reasoning runs at `high` effort by default; if the server rejects the tools + reasoning_effort combo (newer OpenAI models do), postChat drops reasoning_effort for this Client's lifetime so the model still wo

(parent context.Context, messages []chmctx.Message, tools []Tool)

// ProbeResult holds what Probe extracts from a one-shot hello request: the
// live context window and budget snapshot. The TUI shows the real window at
// activation time and feeds the authoritative size to ctx.Pack instead of the
// config.yaml fallback.
type ProbeResult struct {
	ContextWindow int
	Budget        cloud.BudgetStatus
}