chore: add --default-model-provider

This change allow you to set a default model provider where before the
default could only be a url/api pair for OpenAI. Setting this also
implicitly disables the configured openai provider configured with the
--openai* flags

Author: ibuildthecloud

pkg/cli/gptscript.go

@@ -54,24 +54,25 @@ type GPTScript struct {
 	Output         string `usage:"Save output to a file, or - for stdout" short:"o"`
 	EventsStreamTo string `usage:"Stream events to this location, could be a file descriptor/handle (e.g. fd://2), filename, or named pipe (e.g. \\\\.\\pipe\\my-pipe)" name:"events-stream-to"`
 	// Input should not be using GPTSCRIPT_INPUT env var because that is the same value that is set in tool executions
-	Input              string   `usage:"Read input from a file (\"-\" for stdin)" short:"f" env:"GPTSCRIPT_INPUT_FILE"`
-	SubTool            string   `usage:"Use tool of this name, not the first tool in file" local:"true"`
-	Assemble           bool     `usage:"Assemble tool to a single artifact, saved to --output" hidden:"true" local:"true"`
-	ListModels         bool     `usage:"List the models available and exit" local:"true"`
-	ListTools          bool     `usage:"List built-in tools and exit" local:"true"`
-	ListenAddress      string   `usage:"Server listen address" default:"127.0.0.1:0" hidden:"true"`
-	Chdir              string   `usage:"Change current working directory" short:"C"`
-	Daemon             bool     `usage:"Run tool as a daemon" local:"true" hidden:"true"`
-	Ports              string   `usage:"The port range to use for ephemeral daemon ports (ex: 11000-12000)" hidden:"true"`
-	CredentialContext  string   `usage:"Context name in which to store credentials" default:"default"`
-	CredentialOverride []string `usage:"Credentials to override (ex: --credential-override github.com/example/cred-tool:API_TOKEN=1234)"`
-	ChatState          string   `usage:"The chat state to continue, or null to start a new chat and return the state" local:"true"`
-	ForceChat          bool     `usage:"Force an interactive chat session if even the top level tool is not a chat tool" local:"true"`
-	ForceSequential    bool     `usage:"Force parallel calls to run sequentially" local:"true"`
-	Workspace          string   `usage:"Directory to use for the workspace, if specified it will not be deleted on exit"`
-	UI                 bool     `usage:"Launch the UI" local:"true" name:"ui"`
-	DisableTUI         bool     `usage:"Don't use chat TUI but instead verbose output" local:"true" name:"disable-tui"`
-	SaveChatStateFile  string   `usage:"A file to save the chat state to so that a conversation can be resumed with --chat-state" local:"true"`
+	Input                string   `usage:"Read input from a file (\"-\" for stdin)" short:"f" env:"GPTSCRIPT_INPUT_FILE"`
+	SubTool              string   `usage:"Use tool of this name, not the first tool in file" local:"true"`
+	Assemble             bool     `usage:"Assemble tool to a single artifact, saved to --output" hidden:"true" local:"true"`
+	ListModels           bool     `usage:"List the models available and exit" local:"true"`
+	ListTools            bool     `usage:"List built-in tools and exit" local:"true"`
+	ListenAddress        string   `usage:"Server listen address" default:"127.0.0.1:0" hidden:"true"`
+	Chdir                string   `usage:"Change current working directory" short:"C"`
+	Daemon               bool     `usage:"Run tool as a daemon" local:"true" hidden:"true"`
+	Ports                string   `usage:"The port range to use for ephemeral daemon ports (ex: 11000-12000)" hidden:"true"`
+	CredentialContext    string   `usage:"Context name in which to store credentials" default:"default"`
+	CredentialOverride   []string `usage:"Credentials to override (ex: --credential-override github.com/example/cred-tool:API_TOKEN=1234)"`
+	ChatState            string   `usage:"The chat state to continue, or null to start a new chat and return the state" local:"true"`
+	ForceChat            bool     `usage:"Force an interactive chat session if even the top level tool is not a chat tool" local:"true"`
+	ForceSequential      bool     `usage:"Force parallel calls to run sequentially" local:"true"`
+	Workspace            string   `usage:"Directory to use for the workspace, if specified it will not be deleted on exit"`
+	UI                   bool     `usage:"Launch the UI" local:"true" name:"ui"`
+	DisableTUI           bool     `usage:"Don't use chat TUI but instead verbose output" local:"true" name:"disable-tui"`
+	SaveChatStateFile    string   `usage:"A file to save the chat state to so that a conversation can be resumed with --chat-state" local:"true"`
+	DefaultModelProvider string   `usage:"Default LLM model provider to use, this will override OpenAI settings"`
 
 	readData []byte
 }
@@ -136,11 +137,12 @@ func (r *GPTScript) NewGPTScriptOpts() (gptscript.Options, error) {
 			CredentialOverrides: r.CredentialOverride,
 			Sequential:          r.ForceSequential,
 		},
-		Quiet:               r.Quiet,
-		Env:                 os.Environ(),
-		CredentialContext:   r.CredentialContext,
-		Workspace:           r.Workspace,
-		DisablePromptServer: r.UI,
+		Quiet:                r.Quiet,
+		Env:                  os.Environ(),
+		CredentialContext:    r.CredentialContext,
+		Workspace:            r.Workspace,
+		DisablePromptServer:  r.UI,
+		DefaultModelProvider: r.DefaultModelProvider,
 	}
 
 	if r.Confirm {

pkg/gptscript/gptscript.go

@@ -40,15 +40,16 @@ type GPTScript struct {
 }
 
 type Options struct {
-	Cache               cache.Options
-	OpenAI              openai.Options
-	Monitor             monitor.Options
-	Runner              runner.Options
-	CredentialContext   string
-	Quiet               *bool
-	Workspace           string
-	DisablePromptServer bool
-	Env                 []string
+	Cache                cache.Options
+	OpenAI               openai.Options
+	Monitor              monitor.Options
+	Runner               runner.Options
+	DefaultModelProvider string
+	CredentialContext    string
+	Quiet                *bool
+	Workspace            string
+	DisablePromptServer  bool
+	Env                  []string
 }
 
 func Complete(opts ...Options) Options {
@@ -64,6 +65,7 @@ func Complete(opts ...Options) Options {
 		result.Workspace = types.FirstSet(opt.Workspace, result.Workspace)
 		result.Env = append(result.Env, opt.Env...)
 		result.DisablePromptServer = types.FirstSet(opt.DisablePromptServer, result.DisablePromptServer)
+		result.DefaultModelProvider = types.FirstSet(opt.DefaultModelProvider, result.DefaultModelProvider)
 	}
 
 	if result.Quiet == nil {
@@ -106,16 +108,18 @@ func New(ctx context.Context, o ...Options) (*GPTScript, error) {
 		return nil, err
 	}
 
-	oaiClient, err := openai.NewClient(ctx, credStore, opts.OpenAI, openai.Options{
-		Cache:   cacheClient,
-		SetSeed: true,
-	})
-	if err != nil {
-		return nil, err
-	}
+	if opts.DefaultModelProvider == "" {
+		oaiClient, err := openai.NewClient(ctx, credStore, opts.OpenAI, openai.Options{
+			Cache:   cacheClient,
+			SetSeed: true,
+		})
+		if err != nil {
+			return nil, err
+		}
 
-	if err := registry.AddClient(oaiClient); err != nil {
-		return nil, err
+		if err := registry.AddClient(oaiClient); err != nil {
+			return nil, err
+		}
 	}
 
 	if opts.Runner.MonitorFactory == nil {
@@ -143,7 +147,7 @@ func New(ctx context.Context, o ...Options) (*GPTScript, error) {
 
 	fullEnv := append(opts.Env, extraEnv...)
 
-	remoteClient := remote.New(runner, fullEnv, cacheClient, credStore)
+	remoteClient := remote.New(runner, fullEnv, cacheClient, credStore, opts.DefaultModelProvider)
 	if err := registry.AddClient(remoteClient); err != nil {
 		closeServer()
 		return nil, err

pkg/remote/remote.go

@@ -22,21 +22,23 @@ import (
 )
 
 type Client struct {
-	clientsLock sync.Mutex
-	cache       *cache.Client
-	clients     map[string]*openai.Client
-	models      map[string]*openai.Client
-	runner      *runner.Runner
-	envs        []string
-	credStore   credentials.CredentialStore
+	clientsLock     sync.Mutex
+	cache           *cache.Client
+	clients         map[string]*openai.Client
+	models          map[string]*openai.Client
+	runner          *runner.Runner
+	envs            []string
+	credStore       credentials.CredentialStore
+	defaultProvider string
 }
 
-func New(r *runner.Runner, envs []string, cache *cache.Client, credStore credentials.CredentialStore) *Client {
+func New(r *runner.Runner, envs []string, cache *cache.Client, credStore credentials.CredentialStore, defaultProvider string) *Client {
 	return &Client{
-		cache:     cache,
-		runner:    r,
-		envs:      envs,
-		credStore: credStore,
+		cache:           cache,
+		runner:          r,
+		envs:            envs,
+		credStore:       credStore,
+		defaultProvider: defaultProvider,
 	}
 }
 
@@ -73,13 +75,23 @@ func (c *Client) ListModels(ctx context.Context, providers ...string) (result []
 	return
 }
 
-func (c *Client) Supports(ctx context.Context, modelName string) (bool, error) {
-	toolName, modelNameSuffix := types.SplitToolRef(modelName)
-	if modelNameSuffix == "" {
+func (c *Client) parseModel(modelString string) (modelName, providerName string) {
+	toolName, subTool := types.SplitToolRef(modelString)
+	if subTool == "" {
+		// This is just a plain model string "gpt4o"
+		return toolName, c.defaultProvider
+	}
+	// This is a provider string "modelName from provider"
+	return subTool, toolName
+}
+
+func (c *Client) Supports(ctx context.Context, modelString string) (bool, error) {
+	_, providerName := c.parseModel(modelString)
+	if providerName == "" {
 		return false, nil
 	}
 
-	client, err := c.load(ctx, toolName)
+	client, err := c.load(ctx, providerName)
 	if err != nil {
 		return false, err
 	}
@@ -91,7 +103,7 @@ func (c *Client) Supports(ctx context.Context, modelName string) (bool, error) {
 		c.models = map[string]*openai.Client{}
 	}
 
-	c.models[modelName] = client
+	c.models[modelString] = client
 	return true, nil
 }
 

pkg/sdkserver/routes.go

@@ -204,6 +204,7 @@ func (s *server) execHandler(w http.ResponseWriter, r *http.Request) {
 			CredentialOverrides: reqObject.CredentialOverrides,
 			Sequential:          reqObject.ForceSequential,
 		},
+		DefaultModelProvider: reqObject.DefaultModelProvider,
 	}
 
 	if reqObject.Confirm {

pkg/sdkserver/types.go

@@ -52,17 +52,18 @@ type toolOrFileRequest struct {
 	cacheOptions  `json:",inline"`
 	openAIOptions `json:",inline"`
 
-	ToolDefs            toolDefs `json:"toolDefs,inline"`
-	SubTool             string   `json:"subTool"`
-	Input               string   `json:"input"`
-	ChatState           string   `json:"chatState"`
-	Workspace           string   `json:"workspace"`
-	Env                 []string `json:"env"`
-	CredentialContext   string   `json:"credentialContext"`
-	CredentialOverrides []string `json:"credentialOverrides"`
-	Confirm             bool     `json:"confirm"`
-	Location            string   `json:"location,omitempty"`
-	ForceSequential     bool     `json:"forceSequential"`
+	ToolDefs             toolDefs `json:"toolDefs,inline"`
+	SubTool              string   `json:"subTool"`
+	Input                string   `json:"input"`
+	ChatState            string   `json:"chatState"`
+	Workspace            string   `json:"workspace"`
+	Env                  []string `json:"env"`
+	CredentialContext    string   `json:"credentialContext"`
+	CredentialOverrides  []string `json:"credentialOverrides"`
+	Confirm              bool     `json:"confirm"`
+	Location             string   `json:"location,omitempty"`
+	ForceSequential      bool     `json:"forceSequential"`
+	DefaultModelProvider string   `json:"defaultModelProvider,omitempty"`
 }
 
 type content struct {