feat: add managed llm summaries and memory timeline

internal/extraction/generator.go

@@ -18,7 +18,8 @@ import (
 )
 
 type GenerationInput struct {
-	Content string
+	TenantID string
+	Content  string
 }
 
 type GenerationResult struct {
@@ -89,13 +90,21 @@ func safePolicyToken(value string) string {
 }
 
 type ArkGenerator struct {
-	endpoint string
-	apiKey   string
-	model    string
-	client   *http.Client
+	endpoint    string
+	apiKey      string
+	model       string
+	provider    string
+	temperature float64
+	maxTokens   int
+	client      *http.Client
 }
 
 func NewArkGenerator(baseURL, apiKey, model string, timeout time.Duration) (*ArkGenerator, error) {
+	return NewCompatibleGenerator("ark", baseURL, apiKey, model, 0.2, 1200, timeout)
+}
+
+func NewCompatibleGenerator(provider, baseURL, apiKey, model string, temperature float64, maxTokens int, timeout time.Duration) (*ArkGenerator, error) {
+	provider = strings.ToLower(strings.TrimSpace(provider))
 	baseURL = strings.TrimRight(strings.TrimSpace(baseURL), "/")
 	apiKey = strings.TrimSpace(apiKey)
 	model = strings.TrimSpace(model)
@@ -103,11 +112,15 @@ func NewArkGenerator(baseURL, apiKey, model string, timeout time.Duration) (*Ark
 	if err != nil || parsed.Host == "" || (parsed.Scheme != "http" && parsed.Scheme != "https") {
 		return nil, errors.New("Ark base URL must be an absolute HTTP(S) URL")
 	}
-	if apiKey == "" || model == "" || timeout <= 0 {
-		return nil, errors.New("Ark API key, content model and positive timeout are required")
+	if provider != "ark" && provider != "openai_compatible" {
+		return nil, errors.New("content model provider must be ark or openai_compatible")
+	}
+	if apiKey == "" || model == "" || timeout <= 0 || temperature < 0 || temperature > 2 || maxTokens < 256 || maxTokens > 8192 {
+		return nil, errors.New("content model credentials and valid generation parameters are required")
 	}
 	return &ArkGenerator{
 		endpoint: baseURL + "/chat/completions", apiKey: apiKey, model: model,
+		provider: provider, temperature: temperature, maxTokens: maxTokens,
 		client: &http.Client{Timeout: timeout},
 	}, nil
 }
@@ -123,8 +136,8 @@ func (g *ArkGenerator) Generate(ctx context.Context, input GenerationInput) (Gen
 			{"role": "system", "content": "你是长者口述内容整理助手。只依据输入字幕生成可人工核对的候选，不补充未出现的事实，不把推断写成事实。必须返回JSON对象，字段为summary、keywords、open_topic、story_title、story_content、fact_content、confidence。summary概括本次内容，keywords是1至8个简短关键词字符串组成的数组，open_topic给出一个温和的后续话题，story和fact保留原意，confidence取0到1。"},
 			{"role": "user", "content": content},
 		},
-		"temperature":     0.2,
-		"max_tokens":      1200,
+		"temperature":     g.temperature,
+		"max_tokens":      g.maxTokens,
 		"response_format": map[string]string{"type": "json_object"},
 	}
 	encoded, err := json.Marshal(requestBody)
@@ -181,8 +194,8 @@ func (g *ArkGenerator) Generate(ctx context.Context, input GenerationInput) (Gen
 	if !validGeneratedContent(result) {
 		return GenerationResult{}, outbox.Retryable(errors.New("Ark structured content failed validation"))
 	}
-	result.ModelRef = "ark:" + g.model
-	result.PromptVersion = "story-ark-json-v1"
+	result.ModelRef = g.provider + ":" + g.model
+	result.PromptVersion = "story-" + g.provider + "-json-v2"
 	result.AllowStory = true
 	return result, nil
 }