add the greeting msg

go-api/internal/runtime/budget.go

@@ -56,6 +56,15 @@ type Limits struct {
 	MaxToolCalls int
 	MaxTokens    int64
 	Deadline     time.Duration
+
+	// MaxOutputTokens caps a SINGLE model call; MaxTokens caps the whole run.
+	//
+	// Without it the run budget was the only ceiling on any one response, so a
+	// balanced run could spend its 120k as eight 16k generations — and
+	// generation time is the wall clock a person waits through. Latency is why
+	// this exists; cost is a side effect. Zero means the gateway's configured
+	// default (`MODEL_MAX_OUTPUT_TOKENS`).
+	MaxOutputTokens int64
 }
 
 // LimitsForTier is what a run gets when its spec declares no limits of its own.
@@ -70,14 +79,23 @@ type Limits struct {
 // expensive one: a fast run gets a third of a deep run's steps and a sixth of
 // its deadline, so a misrouted spec shows up as a truncated answer rather than
 // as a bill.
+//
+// MaxOutputTokens follows the same shape. It is sized for the longest answer a
+// tier should ever give in one turn, not for the run: a tool-call step spends a
+// few hundred tokens on arguments, and a chat answer past ~3k tokens is one
+// nobody reads. A cap the model is not told about truncates rather than winding
+// down, so these are set above any legitimate answer and not near it.
 func LimitsForTier(tier string) Limits {
 	switch tier {
 	case "fast":
-		return Limits{MaxSteps: 3, MaxToolCalls: 4, MaxTokens: 40_000, Deadline: 20 * time.Second}
+		return Limits{MaxSteps: 3, MaxToolCalls: 4, MaxTokens: 40_000, Deadline: 20 * time.Second,
+			MaxOutputTokens: 1_500}
 	case "deep":
-		return Limits{MaxSteps: 12, MaxToolCalls: 20, MaxTokens: 300_000, Deadline: 120 * time.Second}
+		return Limits{MaxSteps: 12, MaxToolCalls: 20, MaxTokens: 300_000, Deadline: 120 * time.Second,
+			MaxOutputTokens: 4_000}
 	default: // balanced, and anything unrecognised — ParseTier has already normalised it
-		return Limits{MaxSteps: 8, MaxToolCalls: 12, MaxTokens: 120_000, Deadline: 60 * time.Second}
+		return Limits{MaxSteps: 8, MaxToolCalls: 12, MaxTokens: 120_000, Deadline: 60 * time.Second,
+			MaxOutputTokens: 3_000}
 	}
 }