Merge pull request #679 from dgageot/fix-676

Fix model cloning

Author: dgageot

pkg/model/provider/anthropic/client.go

@@ -31,7 +31,7 @@ type Client struct {
 // models:provider_opts:interleaved_thinking: true
 func (c *Client) interleavedThinkingEnabled() bool {
 	// Default to false if not provided
-	if c == nil || c.ModelConfig == nil || len(c.ModelConfig.ProviderOpts) == 0 {
+	if c == nil || len(c.ModelConfig.ProviderOpts) == 0 {
 		return false
 	}
 	v, ok := c.ModelConfig.ProviderOpts["interleaved_thinking"]
@@ -121,14 +121,15 @@ func NewClient(ctx context.Context, cfg *latest.ModelConfig, env environment.Pro
 
 	slog.Debug("Anthropic client created successfully", "model", cfg.Model)
 
-	if globalOptions.StructuredOutput != nil {
-		return &Client{}, errors.New("anthropic does not support native structured_output")
+	if globalOptions.StructuredOutput() != nil {
+		return nil, errors.New("anthropic does not support native structured_output")
 	}
 
 	return &Client{
 		Config: base.Config{
-			ModelConfig:  cfg,
+			ModelConfig:  *cfg,
 			ModelOptions: globalOptions,
+			Env:          env,
 		},
 		clientFn: clientFn,
 	}, nil

pkg/model/provider/base/base.go

@@ -2,30 +2,23 @@ package base
 
 import (
 	latest "github.com/docker/cagent/pkg/config/v2"
+	"github.com/docker/cagent/pkg/environment"
 	"github.com/docker/cagent/pkg/model/provider/options"
 )
 
 // Config is a common base configuration shared by all provider clients.
 // It can be embedded in provider-specific Client structs to avoid code duplication.
 type Config struct {
-	ModelConfig  *latest.ModelConfig
+	ModelConfig  latest.ModelConfig
 	ModelOptions options.ModelOptions
+	Env          environment.Provider
 }
 
 // ID returns the provider and model ID in the format "provider/model"
 func (c *Config) ID() string {
 	return c.ModelConfig.Provider + "/" + c.ModelConfig.Model
 }
 
-// MaxTokens returns the maximum tokens configured for this provider's model
-func (c *Config) MaxTokens() int {
-	if c.ModelConfig == nil {
-		return 0
-	}
-	return c.ModelConfig.MaxTokens
-}
-
-// Options returns the effective model options used by this provider's model
-func (c *Config) Options() options.ModelOptions {
-	return c.ModelOptions
+func (c *Config) BaseConfig() Config {
+	return *c
 }

pkg/model/provider/clone.go

@@ -3,40 +3,25 @@ package provider
 import (
 	"context"
 	"log/slog"
-	"strings"
 
-	latest "github.com/docker/cagent/pkg/config/v2"
-	"github.com/docker/cagent/pkg/environment"
 	"github.com/docker/cagent/pkg/model/provider/options"
 )
 
 // CloneWithOptions returns a new Provider instance using the same provider/model
 // as the base provider, applying the provided options. If cloning fails, the
 // original base provider is returned.
-func CloneWithOptions(ctx context.Context, base Provider, env environment.Provider, opts ...options.Opt) Provider {
-	if base == nil {
-		return nil
-	}
-
-	id := strings.TrimSpace(base.ID())
-	parts := strings.SplitN(id, "/", 2)
-	if len(parts) != 2 {
-		return base
-	}
-
-	cfg := &latest.ModelConfig{Provider: parts[0], Model: parts[1]}
-	if env == nil {
-		env = environment.NewDefaultProvider()
-	}
+func CloneWithOptions(ctx context.Context, base Provider, opts ...options.Opt) Provider {
+	config := base.BaseConfig()
 
 	// Preserve existing options, then apply overrides. Later opts take precedence.
-	baseOpts := options.FromModelOptions(base.Options())
+	baseOpts := options.FromModelOptions(config.ModelOptions)
 	mergedOpts := append(baseOpts, opts...)
 
-	cloned, err := New(ctx, cfg, env, mergedOpts...)
+	clone, err := New(ctx, &config.ModelConfig, config.Env, mergedOpts...)
 	if err != nil {
-		slog.Debug("Failed to clone provider; using base provider", "error", err, "id", id)
+		slog.Debug("Failed to clone provider; using base provider", "error", err, "id", base.ID())
 		return base
 	}
-	return cloned
+
+	return clone
 }

pkg/model/provider/dmr/client.go

@@ -107,7 +107,7 @@ func NewClient(ctx context.Context, cfg *latest.ModelConfig, opts ...options.Opt
 
 	return &Client{
 		Config: base.Config{
-			ModelConfig:  cfg,
+			ModelConfig:  *cfg,
 			ModelOptions: globalOptions,
 		},
 		client:  openai.NewClientWithConfig(clientConfig),
@@ -384,15 +384,16 @@ func (c *Client) CreateChatCompletionStream(ctx context.Context, messages []chat
 	} else {
 		slog.Error("Failed to marshal DMR request to JSON", "error", err)
 	}
-	if c.ModelOptions.StructuredOutput != nil {
-		slog.Debug("Adding structured output to DMR request", "structured_output", c.ModelOptions.StructuredOutput)
+	if structuredOutput := c.ModelOptions.StructuredOutput(); structuredOutput != nil {
+		slog.Debug("Adding structured output to DMR request", "structured_output", structuredOutput)
+
 		request.ResponseFormat = &openai.ChatCompletionResponseFormat{
 			Type: openai.ChatCompletionResponseFormatTypeJSONSchema,
 			JSONSchema: &openai.ChatCompletionResponseFormatJSONSchema{
-				Name:        c.ModelOptions.StructuredOutput.Name,
-				Description: c.ModelOptions.StructuredOutput.Description,
-				Schema:      jsonSchema(c.ModelOptions.StructuredOutput.Schema),
-				Strict:      c.ModelOptions.StructuredOutput.Strict,
+				Name:        structuredOutput.Name,
+				Description: structuredOutput.Description,
+				Schema:      jsonSchema(structuredOutput.Schema),
+				Strict:      structuredOutput.Strict,
 			},
 		}
 	}

pkg/model/provider/gemini/client.go

@@ -37,13 +37,13 @@ func NewClient(ctx context.Context, cfg *latest.ModelConfig, env environment.Pro
 		return nil, errors.New("model type must be 'google'")
 	}
 
-	var modelOptions options.ModelOptions
+	var globalOptions options.ModelOptions
 	for _, opt := range opts {
-		opt(&modelOptions)
+		opt(&globalOptions)
 	}
 
 	var clientFn func(context.Context) (*genai.Client, error)
-	if gateway := modelOptions.Gateway(); gateway == "" {
+	if gateway := globalOptions.Gateway(); gateway == "" {
 		apiKey := env.Get(ctx, "GOOGLE_API_KEY")
 		if apiKey == "" {
 			return nil, errors.New("GOOGLE_API_KEY environment variable is required")
@@ -101,8 +101,9 @@ func NewClient(ctx context.Context, cfg *latest.ModelConfig, env environment.Pro
 
 	return &Client{
 		Config: base.Config{
-			ModelConfig:  cfg,
-			ModelOptions: modelOptions,
+			ModelConfig:  *cfg,
+			ModelOptions: globalOptions,
+			Env:          env,
 		},
 		clientFn: clientFn,
 	}, nil
@@ -213,10 +214,6 @@ func convertMessagesToGemini(messages []chat.Message) []*genai.Content {
 
 // buildConfig creates GenerateContentConfig from model config
 func (c *Client) buildConfig() *genai.GenerateContentConfig {
-	if c.ModelConfig == nil {
-		return nil
-	}
-
 	config := &genai.GenerateContentConfig{
 		Temperature:      genai.Ptr(float32(c.ModelConfig.Temperature)),
 		TopP:             genai.Ptr(float32(c.ModelConfig.TopP)),
@@ -251,9 +248,9 @@ func (c *Client) buildConfig() *genai.GenerateContentConfig {
 		}
 	}
 
-	if c.ModelOptions.StructuredOutput != nil {
+	if structuredOutput := c.ModelOptions.StructuredOutput(); structuredOutput != nil {
 		config.ResponseMIMEType = "application/json"
-		config.ResponseJsonSchema = c.ModelOptions.StructuredOutput.Schema
+		config.ResponseJsonSchema = structuredOutput.Schema
 	}
 
 	return config

pkg/model/provider/openai/client.go

@@ -123,8 +123,9 @@ func NewClient(ctx context.Context, cfg *latest.ModelConfig, env environment.Pro
 
 	return &Client{
 		Config: base.Config{
-			ModelConfig:  cfg,
+			ModelConfig:  *cfg,
 			ModelOptions: globalOptions,
+			Env:          env,
 		},
 		clientFn: clientFn,
 	}, nil
@@ -235,12 +236,12 @@ func (c *Client) CreateChatCompletionStream(
 		},
 	}
 
-	if c.MaxTokens() > 0 {
+	if maxToken := c.ModelConfig.MaxTokens; maxToken > 0 {
 		if !isResponsesOnlyModel(c.ModelConfig.Model) {
-			request.MaxTokens = c.MaxTokens()
-			slog.Debug("OpenAI request configured with max tokens", "max_tokens", c.MaxTokens())
+			request.MaxTokens = maxToken
+			slog.Debug("OpenAI request configured with max tokens", "max_tokens", maxToken, "model", c.ModelConfig.Model)
 		} else {
-			request.MaxCompletionTokens = c.MaxTokens()
+			request.MaxCompletionTokens = maxToken
 			slog.Debug("using max_completion_tokens instead of max_tokens for Responses-API models", "model", c.ModelConfig.Model)
 		}
 	}
@@ -273,7 +274,7 @@ func (c *Client) CreateChatCompletionStream(
 
 	// Apply thinking budget: set reasoning_effort parameter
 	if c.ModelConfig.ThinkingBudget != nil {
-		effort, err := getOpenAIReasoningEffort(c.ModelConfig)
+		effort, err := getOpenAIReasoningEffort(&c.ModelConfig)
 		if err != nil {
 			slog.Error("OpenAI request using thinking_budget failed", "error", err)
 			return nil, err
@@ -283,17 +284,18 @@ func (c *Client) CreateChatCompletionStream(
 	}
 
 	// Apply structured output configuration
-	if c.ModelOptions.StructuredOutput != nil {
+	if structuredOutput := c.ModelOptions.StructuredOutput(); structuredOutput != nil {
+		slog.Debug("OpenAI request using structured output", "name", structuredOutput.Name, "strict", structuredOutput.Strict)
+
 		request.ResponseFormat = &openai.ChatCompletionResponseFormat{
 			Type: openai.ChatCompletionResponseFormatTypeJSONSchema,
 			JSONSchema: &openai.ChatCompletionResponseFormatJSONSchema{
-				Name:        c.ModelOptions.StructuredOutput.Name,
-				Description: c.ModelOptions.StructuredOutput.Description,
-				Schema:      jsonSchema(c.ModelOptions.StructuredOutput.Schema),
-				Strict:      c.ModelOptions.StructuredOutput.Strict,
+				Name:        structuredOutput.Name,
+				Description: structuredOutput.Description,
+				Schema:      jsonSchema(structuredOutput.Schema),
+				Strict:      structuredOutput.Strict,
 			},
 		}
-		slog.Debug("OpenAI request using structured output", "name", c.ModelOptions.StructuredOutput.Name, "strict", c.ModelOptions.StructuredOutput.Strict)
 	}
 
 	// Log the request in JSON format for debugging

pkg/model/provider/options/options.go

@@ -6,13 +6,17 @@ import (
 
 type ModelOptions struct {
 	gateway          string
-	StructuredOutput *latest.StructuredOutput
+	structuredOutput *latest.StructuredOutput
 }
 
 func (c *ModelOptions) Gateway() string {
 	return c.gateway
 }
 
+func (c *ModelOptions) StructuredOutput() *latest.StructuredOutput {
+	return c.structuredOutput
+}
+
 type Opt func(*ModelOptions)
 
 func WithGateway(gateway string) Opt {
@@ -21,9 +25,9 @@ func WithGateway(gateway string) Opt {
 	}
 }
 
-func WithStructuredOutput(output *latest.StructuredOutput) Opt {
+func WithStructuredOutput(structuredOutput *latest.StructuredOutput) Opt {
 	return func(cfg *ModelOptions) {
-		cfg.StructuredOutput = output
+		cfg.structuredOutput = structuredOutput
 	}
 }
 
@@ -34,8 +38,8 @@ func FromModelOptions(m ModelOptions) []Opt {
 	if g := m.Gateway(); g != "" {
 		out = append(out, WithGateway(g))
 	}
-	if m.StructuredOutput != nil {
-		out = append(out, WithStructuredOutput(m.StructuredOutput))
+	if m.structuredOutput != nil {
+		out = append(out, WithStructuredOutput(m.structuredOutput))
 	}
 	return out
 }

pkg/model/provider/provider.go

@@ -9,6 +9,7 @@ import (
 	latest "github.com/docker/cagent/pkg/config/v2"
 	"github.com/docker/cagent/pkg/environment"
 	"github.com/docker/cagent/pkg/model/provider/anthropic"
+	"github.com/docker/cagent/pkg/model/provider/base"
 	"github.com/docker/cagent/pkg/model/provider/dmr"
 	"github.com/docker/cagent/pkg/model/provider/gemini"
 	"github.com/docker/cagent/pkg/model/provider/openai"
@@ -52,10 +53,8 @@ type Provider interface {
 		messages []chat.Message,
 		tools []tools.Tool,
 	) (chat.MessageStream, error)
-	// Options returns the effective model options used by this provider
-	Options() options.ModelOptions
-	// MaxTokens returns the maximum tokens configured for this provider
-	MaxTokens() int
+	// BaseConfig returns the base configuration of this provider
+	BaseConfig() base.Config
 }
 
 func New(ctx context.Context, cfg *latest.ModelConfig, env environment.Provider, opts ...options.Opt) (Provider, error) {

pkg/runtime/runtime.go

@@ -1014,7 +1014,7 @@ func (r *LocalRuntime) generateSessionTitle(ctx context.Context, sess *session.S
 	systemPrompt := "You are a helpful AI assistant that generates concise, descriptive titles for conversations. You will be given a conversation history and asked to create a title that captures the main topic."
 	userPrompt := fmt.Sprintf("Based on the following message a user sent to an AI assistant, generate a short, descriptive title (maximum 50 characters) that captures the main topic or purpose of the conversation. Return ONLY the title text, nothing else.\n\nUser message:%s\n\n", conversationHistory.String())
 
-	titleModel := provider.CloneWithOptions(ctx, r.CurrentAgent().Model(), nil, options.WithStructuredOutput(nil))
+	titleModel := provider.CloneWithOptions(ctx, r.CurrentAgent().Model(), options.WithStructuredOutput(nil))
 	newTeam := team.New(
 		team.WithID("title-generator"),
 		team.WithAgents(agent.New("root", systemPrompt, agent.WithModel(titleModel))),
@@ -1080,7 +1080,7 @@ func (r *LocalRuntime) Summarize(ctx context.Context, sess *session.Session, eve
 	// Create a new session for summary generation
 	systemPrompt := "You are a helpful AI assistant that creates comprehensive summaries of conversations. You will be given a conversation history and asked to create a concise yet thorough summary that captures the key points, decisions made, and outcomes."
 	userPrompt := fmt.Sprintf("Based on the following conversation between a user and an AI assistant, create a comprehensive summary that captures:\n- The main topics discussed\n- Key information exchanged\n- Decisions made or conclusions reached\n- Important outcomes or results\n\nProvide a well-structured summary (2-4 paragraphs) that someone could read to understand what happened in this conversation. Return ONLY the summary text, nothing else.\n\nConversation history:%s\n\nGenerate a summary for this conversation:", conversationHistory.String())
-	newModel := provider.CloneWithOptions(ctx, r.CurrentAgent().Model(), nil, options.WithStructuredOutput(nil))
+	newModel := provider.CloneWithOptions(ctx, r.CurrentAgent().Model(), options.WithStructuredOutput(nil))
 	newTeam := team.New(
 		team.WithID("summary-generator"),
 		team.WithAgents(agent.New("root", systemPrompt, agent.WithModel(newModel))),

pkg/runtime/runtime_test.go

@@ -13,7 +13,7 @@ import (
 
 	"github.com/docker/cagent/pkg/agent"
 	"github.com/docker/cagent/pkg/chat"
-	"github.com/docker/cagent/pkg/model/provider/options"
+	"github.com/docker/cagent/pkg/model/provider/base"
 	"github.com/docker/cagent/pkg/modelsdev"
 	"github.com/docker/cagent/pkg/session"
 	"github.com/docker/cagent/pkg/team"
@@ -143,7 +143,7 @@ func (m *mockProvider) CreateChatCompletionStream(context.Context, []chat.Messag
 	return m.stream, nil
 }
 
-func (m *mockProvider) Options() options.ModelOptions { return options.ModelOptions{} }
+func (m *mockProvider) BaseConfig() base.Config { return base.Config{} }
 
 func (m *mockProvider) MaxTokens() int { return 0 }
 
@@ -157,7 +157,7 @@ func (m *mockProviderWithError) CreateChatCompletionStream(context.Context, []ch
 	return nil, fmt.Errorf("simulated error creating chat completion stream")
 }
 
-func (m *mockProviderWithError) Options() options.ModelOptions { return options.ModelOptions{} }
+func (m *mockProviderWithError) BaseConfig() base.Config { return base.Config{} }
 
 func (m *mockProviderWithError) MaxTokens() int { return 0 }
 
@@ -453,7 +453,7 @@ func (p *queueProvider) CreateChatCompletionStream(context.Context, []chat.Messa
 	return s, nil
 }
 
-func (p *queueProvider) Options() options.ModelOptions { return options.ModelOptions{} }
+func (p *queueProvider) BaseConfig() base.Config { return base.Config{} }
 
 func (p *queueProvider) MaxTokens() int { return 0 }