feat: add ability to pass request-specific env vars to chat completion

This will allow authentication per-request in model providers.

Signed-off-by: Donnie Adams <donnie@acorn.io>

Author: thedadams

go.mod

@@ -15,7 +15,7 @@ require (
 	github.com/google/shlex v0.0.0-20191202100458-e7afc7fbc510
 	github.com/google/uuid v1.6.0
 	github.com/gptscript-ai/broadcaster v0.0.0-20240625175512-c43682019b86
-	github.com/gptscript-ai/chat-completion-client v0.0.0-20240813051153-a440ada7e3c3
+	github.com/gptscript-ai/chat-completion-client v0.0.0-20241104122544-5fe75f07c131
 	github.com/gptscript-ai/cmd v0.0.0-20240802230653-326b7baf6fcb
 	github.com/gptscript-ai/go-gptscript v0.9.5-rc5.0.20240927213153-2af51434b93e
 	github.com/gptscript-ai/tui v0.0.0-20240923192013-172e51ccf1d6

go.sum

@@ -200,8 +200,8 @@ github.com/gorilla/css v1.0.0 h1:BQqNyPTi50JCFMTw/b67hByjMVXZRwGha6wxVGkeihY=
 github.com/gorilla/css v1.0.0/go.mod h1:Dn721qIggHpt4+EFCcTLTU/vk5ySda2ReITrtgBl60c=
 github.com/gptscript-ai/broadcaster v0.0.0-20240625175512-c43682019b86 h1:m9yLtIEd0z1ia8qFjq3u0Ozb6QKwidyL856JLJp6nbA=
 github.com/gptscript-ai/broadcaster v0.0.0-20240625175512-c43682019b86/go.mod h1:lK3K5EZx4dyT24UG3yCt0wmspkYqrj4D/8kxdN3relk=
-github.com/gptscript-ai/chat-completion-client v0.0.0-20240813051153-a440ada7e3c3 h1:EQiFTZv+BnOWJX2B9XdF09fL2Zj7h19n1l23TpWCafc=
-github.com/gptscript-ai/chat-completion-client v0.0.0-20240813051153-a440ada7e3c3/go.mod h1:7P/o6/IWa1KqsntVf68hSnLKuu3+xuqm6lYhch1w4jo=
+github.com/gptscript-ai/chat-completion-client v0.0.0-20241104122544-5fe75f07c131 h1:y2FcmT4X8U606gUS0teX5+JWX9K/NclsLEhHiyrd+EU=
+github.com/gptscript-ai/chat-completion-client v0.0.0-20241104122544-5fe75f07c131/go.mod h1:7P/o6/IWa1KqsntVf68hSnLKuu3+xuqm6lYhch1w4jo=
 github.com/gptscript-ai/cmd v0.0.0-20240802230653-326b7baf6fcb h1:ky2J2CzBOskC7Jgm2VJAQi2x3p7FVGa+2/PcywkFJuc=
 github.com/gptscript-ai/cmd v0.0.0-20240802230653-326b7baf6fcb/go.mod h1:DJAo1xTht1LDkNYFNydVjTHd576TC7MlpsVRl3oloVw=
 github.com/gptscript-ai/go-gptscript v0.9.5-rc5.0.20240927213153-2af51434b93e h1:WpNae0NBx+Ri8RB3SxF8DhadDKU7h+jfWPQterDpbJA=

pkg/context/context.go

@@ -46,14 +46,3 @@ func GetLogger(ctx context.Context) mvl.Logger {
 
 	return l
 }
-
-type envKey struct{}
-
-func WithEnv(ctx context.Context, env []string) context.Context {
-	return context.WithValue(ctx, envKey{}, env)
-}
-
-func GetEnv(ctx context.Context) []string {
-	l, _ := ctx.Value(envKey{}).([]string)
-	return l
-}

pkg/engine/engine.go

@@ -8,7 +8,6 @@ import (
 	"sync"
 
 	"github.com/gptscript-ai/gptscript/pkg/config"
-	gcontext "github.com/gptscript-ai/gptscript/pkg/context"
 	"github.com/gptscript-ai/gptscript/pkg/counter"
 	"github.com/gptscript-ai/gptscript/pkg/types"
 	"github.com/gptscript-ai/gptscript/pkg/version"
@@ -389,7 +388,8 @@ func (e *Engine) complete(ctx context.Context, state *State) (*Return, error) {
 		}
 	}()
 
-	resp, err := e.Model.Call(gcontext.WithEnv(ctx, e.Env), state.Completion, progress)
+	state.Completion.Env = e.Env
+	resp, err := e.Model.Call(ctx, state.Completion, progress)
 	if err != nil {
 		return nil, err
 	}

pkg/llm/proxy.go

@@ -54,7 +54,7 @@ func (r *Registry) ServeHTTP(w http.ResponseWriter, req *http.Request) {
 
 	var (
 		model string
-		data  = map[string]any{}
+		data  map[string]any
 	)
 
 	if json.Unmarshal(inBytes, &data) == nil {
@@ -65,7 +65,7 @@ func (r *Registry) ServeHTTP(w http.ResponseWriter, req *http.Request) {
 		model = builtin.GetDefaultModel()
 	}
 
-	c, err := r.getClient(req.Context(), model)
+	c, err := r.getClient(req.Context(), model, nil)
 	if err != nil {
 		http.Error(w, err.Error(), http.StatusInternalServerError)
 		return

pkg/llm/registry.go

@@ -78,7 +78,7 @@ func (r *Registry) fastPath(modelName string) Client {
 	return r.clients[0]
 }
 
-func (r *Registry) getClient(ctx context.Context, modelName string) (Client, error) {
+func (r *Registry) getClient(ctx context.Context, modelName string, env []string) (Client, error) {
 	if c := r.fastPath(modelName); c != nil {
 		return c, nil
 	}
@@ -101,7 +101,7 @@ func (r *Registry) getClient(ctx context.Context, modelName string) (Client, err
 
 	if len(errs) > 0 && oaiClient != nil {
 		// Prompt the user to enter their OpenAI API key and try again.
-		if err := oaiClient.RetrieveAPIKey(ctx); err != nil {
+		if err := oaiClient.RetrieveAPIKey(ctx, env); err != nil {
 			return nil, err
 		}
 		ok, err := oaiClient.Supports(ctx, modelName)
@@ -146,7 +146,7 @@ func (r *Registry) Call(ctx context.Context, messageRequest types.CompletionRequ
 
 	if len(errs) > 0 && oaiClient != nil {
 		// Prompt the user to enter their OpenAI API key and try again.
-		if err := oaiClient.RetrieveAPIKey(ctx); err != nil {
+		if err := oaiClient.RetrieveAPIKey(ctx, messageRequest.Env); err != nil {
 			return nil, err
 		}
 		ok, err := oaiClient.Supports(ctx, messageRequest.Model)

pkg/openai/client.go

@@ -13,7 +13,6 @@ import (
 
 	openai "github.com/gptscript-ai/chat-completion-client"
 	"github.com/gptscript-ai/gptscript/pkg/cache"
-	gcontext "github.com/gptscript-ai/gptscript/pkg/context"
 	"github.com/gptscript-ai/gptscript/pkg/counter"
 	"github.com/gptscript-ai/gptscript/pkg/credentials"
 	"github.com/gptscript-ai/gptscript/pkg/hash"
@@ -305,7 +304,7 @@ func toMessages(request types.CompletionRequest, compat bool) (result []openai.C
 
 func (c *Client) Call(ctx context.Context, messageRequest types.CompletionRequest, status chan<- types.CompletionStatus) (*types.CompletionMessage, error) {
 	if err := c.ValidAuth(); err != nil {
-		if err := c.RetrieveAPIKey(ctx); err != nil {
+		if err := c.RetrieveAPIKey(ctx, messageRequest.Env); err != nil {
 			return nil, err
 		}
 	}
@@ -401,15 +400,15 @@ func (c *Client) Call(ctx context.Context, messageRequest types.CompletionReques
 	if err != nil {
 		return nil, err
 	} else if !ok {
-		result, err = c.call(ctx, request, id, status)
+		result, err = c.call(ctx, request, id, messageRequest.Env, status)
 
 		// If we got back a context length exceeded error, keep retrying and shrinking the message history until we pass.
 		var apiError *openai.APIError
 		if errors.As(err, &apiError) && apiError.Code == "context_length_exceeded" && messageRequest.Chat {
 			// Decrease maxTokens by 10% to make garbage collection more aggressive.
 			// The retry loop will further decrease maxTokens if needed.
 			maxTokens := decreaseTenPercent(messageRequest.MaxTokens)
-			result, err = c.contextLimitRetryLoop(ctx, request, id, maxTokens, status)
+			result, err = c.contextLimitRetryLoop(ctx, request, id, messageRequest.Env, maxTokens, status)
 		}
 		if err != nil {
 			return nil, err
@@ -443,7 +442,7 @@ func (c *Client) Call(ctx context.Context, messageRequest types.CompletionReques
 	return &result, nil
 }
 
-func (c *Client) contextLimitRetryLoop(ctx context.Context, request openai.ChatCompletionRequest, id string, maxTokens int, status chan<- types.CompletionStatus) (types.CompletionMessage, error) {
+func (c *Client) contextLimitRetryLoop(ctx context.Context, request openai.ChatCompletionRequest, id string, env []string, maxTokens int, status chan<- types.CompletionStatus) (types.CompletionMessage, error) {
 	var (
 		response types.CompletionMessage
 		err      error
@@ -452,7 +451,7 @@ func (c *Client) contextLimitRetryLoop(ctx context.Context, request openai.ChatC
 	for range 10 { // maximum 10 tries
 		// Try to drop older messages again, with a decreased max tokens.
 		request.Messages = dropMessagesOverCount(maxTokens, request.Messages)
-		response, err = c.call(ctx, request, id, status)
+		response, err = c.call(ctx, request, id, env, status)
 		if err == nil {
 			return response, nil
 		}
@@ -542,7 +541,7 @@ func override(left, right string) string {
 	return left
 }
 
-func (c *Client) call(ctx context.Context, request openai.ChatCompletionRequest, transactionID string, partial chan<- types.CompletionStatus) (types.CompletionMessage, error) {
+func (c *Client) call(ctx context.Context, request openai.ChatCompletionRequest, transactionID string, env []string, partial chan<- types.CompletionStatus) (types.CompletionMessage, error) {
 	streamResponse := os.Getenv("GPTSCRIPT_INTERNAL_OPENAI_STREAMING") != "false"
 
 	partial <- types.CompletionStatus{
@@ -553,11 +552,27 @@ func (c *Client) call(ctx context.Context, request openai.ChatCompletionRequest,
 		},
 	}
 
+	var (
+		headers          map[string]string
+		modelProviderEnv []string
+	)
+	for _, e := range env {
+		if strings.HasPrefix(e, "GPTSCRIPT_MODEL_PROVIDER_") {
+			modelProviderEnv = append(modelProviderEnv, e)
+		}
+	}
+
+	if len(modelProviderEnv) > 0 {
+		headers = map[string]string{
+			"X-GPTScript-Env": strings.Join(modelProviderEnv, ","),
+		}
+	}
+
 	slog.Debug("calling openai", "message", request.Messages)
 
 	if !streamResponse {
 		request.StreamOptions = nil
-		resp, err := c.c.CreateChatCompletion(ctx, request)
+		resp, err := c.c.CreateChatCompletion(ctx, request, headers)
 		if err != nil {
 			return types.CompletionMessage{}, err
 		}
@@ -582,7 +597,7 @@ func (c *Client) call(ctx context.Context, request openai.ChatCompletionRequest,
 		}), nil
 	}
 
-	stream, err := c.c.CreateChatCompletionStream(ctx, request)
+	stream, err := c.c.CreateChatCompletionStream(ctx, request, headers)
 	if err != nil {
 		return types.CompletionMessage{}, err
 	}
@@ -614,8 +629,8 @@ func (c *Client) call(ctx context.Context, request openai.ChatCompletionRequest,
 	}
 }
 
-func (c *Client) RetrieveAPIKey(ctx context.Context) error {
-	k, err := prompt.GetModelProviderCredential(ctx, c.credStore, BuiltinCredName, "OPENAI_API_KEY", "Please provide your OpenAI API key:", gcontext.GetEnv(ctx))
+func (c *Client) RetrieveAPIKey(ctx context.Context, env []string) error {
+	k, err := prompt.GetModelProviderCredential(ctx, c.credStore, BuiltinCredName, "OPENAI_API_KEY", "Please provide your OpenAI API key:", env)
 	if err != nil {
 		return err
 	}

pkg/remote/remote.go

@@ -10,7 +10,6 @@ import (
 	"sync"
 
 	"github.com/gptscript-ai/gptscript/pkg/cache"
-	gcontext "github.com/gptscript-ai/gptscript/pkg/context"
 	"github.com/gptscript-ai/gptscript/pkg/credentials"
 	"github.com/gptscript-ai/gptscript/pkg/engine"
 	env2 "github.com/gptscript-ai/gptscript/pkg/env"
@@ -48,7 +47,7 @@ func (c *Client) Call(ctx context.Context, messageRequest types.CompletionReques
 		return nil, fmt.Errorf("failed to find remote model %s", messageRequest.Model)
 	}
 
-	client, err := c.load(ctx, provider)
+	client, err := c.load(ctx, provider, messageRequest.Env...)
 	if err != nil {
 		return nil, err
 	}
@@ -111,7 +110,7 @@ func isHTTPURL(toolName string) bool {
 		strings.HasPrefix(toolName, "https://")
 }
 
-func (c *Client) clientFromURL(ctx context.Context, apiURL string) (*openai.Client, error) {
+func (c *Client) clientFromURL(ctx context.Context, apiURL string, envs []string) (*openai.Client, error) {
 	parsed, err := url.Parse(apiURL)
 	if err != nil {
 		return nil, err
@@ -121,7 +120,7 @@ func (c *Client) clientFromURL(ctx context.Context, apiURL string) (*openai.Clie
 
 	if key == "" && !isLocalhost(apiURL) {
 		var err error
-		key, err = c.retrieveAPIKey(ctx, env, apiURL)
+		key, err = c.retrieveAPIKey(ctx, env, apiURL, envs)
 		if err != nil {
 			return nil, err
 		}
@@ -134,7 +133,7 @@ func (c *Client) clientFromURL(ctx context.Context, apiURL string) (*openai.Clie
 	})
 }
 
-func (c *Client) load(ctx context.Context, toolName string) (*openai.Client, error) {
+func (c *Client) load(ctx context.Context, toolName string, env ...string) (*openai.Client, error) {
 	c.clientsLock.Lock()
 	defer c.clientsLock.Unlock()
 
@@ -144,7 +143,7 @@ func (c *Client) load(ctx context.Context, toolName string) (*openai.Client, err
 	}
 
 	if isHTTPURL(toolName) {
-		remoteClient, err := c.clientFromURL(ctx, toolName)
+		remoteClient, err := c.clientFromURL(ctx, toolName, env)
 		if err != nil {
 			return nil, err
 		}
@@ -183,8 +182,8 @@ func (c *Client) load(ctx context.Context, toolName string) (*openai.Client, err
 	return oClient, nil
 }
 
-func (c *Client) retrieveAPIKey(ctx context.Context, env, url string) (string, error) {
-	return prompt.GetModelProviderCredential(ctx, c.credStore, url, env, fmt.Sprintf("Please provide your API key for %s", url), append(gcontext.GetEnv(ctx), c.envs...))
+func (c *Client) retrieveAPIKey(ctx context.Context, env, url string, envs []string) (string, error) {
+	return prompt.GetModelProviderCredential(ctx, c.credStore, url, env, fmt.Sprintf("Please provide your API key for %s", url), append(envs, c.envs...))
 }
 
 func isLocalhost(url string) bool {

pkg/runner/output.go

@@ -84,7 +84,7 @@ func (r *Runner) handleOutput(callCtx engine.Context, monitor Monitor, env []str
 		if err != nil {
 			return nil, fmt.Errorf("marshaling input for output filter: %w", err)
 		}
-		res, err := r.subCall(callCtx.Ctx, callCtx, monitor, env, outputToolRef.ToolID, string(inputData), "", engine.OutputToolCategory)
+		res, err := r.subCall(callCtx.Ctx, callCtx, monitor, env, outputToolRef.ToolID, inputData, "", engine.OutputToolCategory)
 		if err != nil {
 			return nil, err
 		}

pkg/tests/judge/judge.go

@@ -112,7 +112,7 @@ func (j *Judge[T]) Equal(ctx context.Context, expected, actual T, criteria strin
 			},
 		},
 	}
-	response, err := j.client.CreateChatCompletion(ctx, request)
+	response, err := j.client.CreateChatCompletion(ctx, request, nil)
 	if err != nil {
 		return false, "", fmt.Errorf("failed to create chat completion request: %w", err)
 	}