From c40b46b66182d3c2901b1cae14afe64e99ba735f Mon Sep 17 00:00:00 2001
From: Mathias <mthbqv@gmail.com>
Date: Tue, 2 Jun 2026 20:55:49 +0200
Subject: [PATCH] fix(llm): send generous max_tokens on every request

The copied OpenAI-compatible client sent no max_tokens. Thinking models
(qwen3, deepseek-r1) spend their budget on the reasoning trace and return
EMPTY content when max_tokens is unset, which the summarizer treats as an
error. ADR-004 says change Tapir's copy rather than the hyperguild upstream,
so set a generous default (8192) leaving room for both reasoning and output.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 internal/adapters/llm/client.go      | 12 ++++++++++++
 internal/adapters/llm/client_test.go | 21 +++++++++++++++++++++
 2 files changed, 33 insertions(+)

diff --git a/internal/adapters/llm/client.go b/internal/adapters/llm/client.go
index 0c84a36..8e40af7 100644
--- a/internal/adapters/llm/client.go
+++ b/internal/adapters/llm/client.go
@@ -17,11 +17,20 @@ import (
 	"time"
 )
 
+// defaultMaxTokens is sent on every request. Tapir CHANGES this from the
+// hyperguild copy (ADR-004 says change the copy, not the upstream): thinking
+// models (qwen3, deepseek-r1) spend their budget on reasoning and return EMPTY
+// content when max_tokens is unset or too low. A generous ceiling leaves room
+// for both the reasoning trace and the actual summary. See
+// docs/homelab-integration.md.
+const defaultMaxTokens = 8192
+
 // Client calls an OpenAI-compatible chat completions endpoint.
 type Client struct {
 	baseURL    string
 	apiKey     string
 	model      string
+	maxTokens  int
 	httpClient *http.Client
 }
 
@@ -31,6 +40,7 @@ func New(baseURL, apiKey, model string, timeout time.Duration) *Client {
 		baseURL:    strings.TrimRight(baseURL, "/"),
 		apiKey:     apiKey,
 		model:      model,
+		maxTokens:  defaultMaxTokens,
 		httpClient: &http.Client{Timeout: timeout},
 	}
 }
@@ -39,6 +49,7 @@ type chatRequest struct {
 	Model       string    `json:"model"`
 	Messages    []message `json:"messages"`
 	Temperature float64   `json:"temperature"`
+	MaxTokens   int       `json:"max_tokens,omitempty"`
 }
 
 type message struct {
@@ -62,6 +73,7 @@ func (c *Client) Complete(ctx context.Context, system, user string) (string, err
 			{Role: "user", Content: user},
 		},
 		Temperature: 0.2,
+		MaxTokens:   c.maxTokens,
 	}
 	b, err := json.Marshal(body)
 	if err != nil {
diff --git a/internal/adapters/llm/client_test.go b/internal/adapters/llm/client_test.go
index 742bc96..1f0dc91 100644
--- a/internal/adapters/llm/client_test.go
+++ b/internal/adapters/llm/client_test.go
@@ -43,6 +43,27 @@ func TestClient_Complete(t *testing.T) {
 	}
 }
 
+// TestClient_SendsMaxTokens guards Tapir's ADR-004 change to the copied client:
+// it MUST send a positive max_tokens, or thinking models return empty content.
+func TestClient_SendsMaxTokens(t *testing.T) {
+	var body chatRequest
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		_ = json.NewDecoder(r.Body).Decode(&body)
+		_ = json.NewEncoder(w).Encode(map[string]any{
+			"choices": []map[string]any{{"message": map[string]any{"content": "ok"}}},
+		})
+	}))
+	defer srv.Close()
+
+	c := New(srv.URL, "", "test-model", 10*time.Second)
+	if _, err := c.Complete(context.Background(), "sys", "user"); err != nil {
+		t.Fatalf("Complete: %v", err)
+	}
+	if body.MaxTokens <= 0 {
+		t.Errorf("max_tokens = %d, want > 0 (thinking models return empty content without it)", body.MaxTokens)
+	}
+}
+
 func TestClient_ReturnsErrorOnNon200(t *testing.T) {
 	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 		http.Error(w, "overloaded", http.StatusServiceUnavailable)