Add --max-tokens param to cagent new to allow manual context length overriding

Signed-off-by: Christopher Petito <chrisjpetito@gmail.com>

Author: krissetto

README.md

@@ -208,22 +208,28 @@ To use the feature, you must have an Anthropic, OpenAI or Google API key availab
 
 You can choose what provider and model gets used by passing the `--model provider/modelname` flag to `cagent new`
 
-If `--model` is unspecified, `cagent new` will automatically choose between these 3 providers in order based on the first api key it finds in your environment
+If `--model` is unspecified, `cagent new` will automatically choose between these 3 providers in order based on the first api key it finds in your environment.
 
 ```sh
 export ANTHROPIC_API_KEY=your_api_key_here  # first choice. default model claude-sonnet-4-0
 export OPENAI_API_KEY=your_api_key_here     # if anthropic key not set. default model gpt-5-mini
 export GOOGLE_API_KEY=your_api_key_here     # if anthropic and openai keys are not set. default model gemini-2.5-flash
 ```
 
-Example of provider and model overriding:
+`--max-tokens` can be specified to override the context limit used.  
+When using DMR, the default is 16k to limit memory usage. With all other providers the default is 64k
+
+Example of provider, model and context size overriding:
 
 ```sh
 # Use GPT-5 via OpenAI
 cagent new --model openai/gpt-5
 
 # Use a local model (ai/gemma3-qat:12B) via DMR
 cagent new --model dmr/ai/gemma3-qat:12B
+
+# Override the max_tokens used during generation, default is 64k, 16k when using the dmr provider
+cagent new --model openai/gpt-5-mini --max-tokens 32000
 ```
 
 ---

cmd/root/new.go

@@ -13,7 +13,10 @@ import (
 	"github.com/docker/cagent/pkg/runtime"
 )
 
-var modelParam string
+var (
+	modelParam     string
+	maxTokensParam int
+)
 
 // Cmd creates a new command to create a new agent configuration
 func NewNewCmd() *cobra.Command {
@@ -90,7 +93,7 @@ func NewNewCmd() *cobra.Command {
 				fmt.Println()
 			}
 
-			out, err := creator.StreamCreateAgent(ctx, ".", prompt, runConfig, modelProvider, model)
+			out, err := creator.StreamCreateAgent(ctx, ".", prompt, runConfig, modelProvider, model, maxTokensParam)
 			if err != nil {
 				return err
 			}
@@ -131,6 +134,7 @@ func NewNewCmd() *cobra.Command {
 	}
 	addGatewayFlags(cmd)
 	cmd.PersistentFlags().StringVar(&modelParam, "model", "", "Model to use, optionally as provider/model where provider is one of: anthropic, openai, google, dmr. If omitted, provider is auto-selected based on available credentials or gateway")
+	cmd.PersistentFlags().IntVar(&maxTokensParam, "max-tokens", 0, "Override max_tokens for the selected model (0 = default)")
 
 	return cmd
 }

docs/USAGE.md

@@ -50,6 +50,7 @@ $ cagent api config.yaml --port 8080
 
 # Project Management
 $ cagent new                          # Initialize new project
+$ cagent new --model openai/gpt-5-mini --max-tokens 32000  # Override max tokens during generation
 $ cagent eval config.yaml             # Run evaluations
 $ cagent pull docker.io/user/agent    # Pull agent from registry
 $ cagent push docker.io/user/agent    # Push agent to registry

internal/creator/agent.go

@@ -120,7 +120,7 @@ func CreateAgent(ctx context.Context, baseDir, prompt string, runConfig latest.R
 	return messages[len(messages)-1].Message.Content, fsToolset.path, nil
 }
 
-func StreamCreateAgent(ctx context.Context, baseDir, prompt string, runConfig latest.RuntimeConfig, providerName, modelNameOverride string) (<-chan runtime.Event, error) {
+func StreamCreateAgent(ctx context.Context, baseDir, prompt string, runConfig latest.RuntimeConfig, providerName, modelNameOverride string, maxTokensOverride int) (<-chan runtime.Event, error) {
 	defaultModels := map[string]string{
 		"openai":    "gpt-5-mini",
 		"anthropic": "claude-sonnet-4-0",
@@ -168,6 +168,9 @@ func StreamCreateAgent(ctx context.Context, baseDir, prompt string, runConfig la
 	if providerName == "dmr" {
 		maxTokens = 16000
 	}
+	if maxTokensOverride > 0 {
+		maxTokens = maxTokensOverride
+	}
 
 	llm, err := provider.New(
 		ctx,