Tools fix

CHANGELOG.md

@@ -1,5 +1,15 @@
 # Changelog
 
+# 0.13.5
+- Option to append template to the end in `getOpenAIBody`.
+- Fix IntelliSense for tools when using Vercel AI SDK.
+
+# 0.13.4
+- add parallelToolCalls option to pass it to the API
+
+# 0.13.3
+- Support for Langtail hosted tools
+
 # 0.5.4
 
 - Fix: don't send `stop` parameter if it's empty (this causes validation error in OpenAI in some cases)

README.md

@@ -203,6 +203,9 @@ const joke = await openai.chat.completions.create(openAiBody)
 
 This way you are still using langtail prompts without exposing potentially sensitive data in your variables.
 
+## Typed inputs
+
+You can override input types to improve IntelliSense for the `prompt`, `environment`, `version` and `variables` when calling a prompt. Use the command `npx langtail generate-types`.
 
 ## Vercel AI provider
 
@@ -249,7 +252,7 @@ const result = await generateText({
 
 ### Using tools from Langtail
 
-If your prompts in Langtail contain tools, you can generate a file containing tool parameters for every prompt deployment in your project. Run `npx langtail generate-tools --out [output_filepath]` to generate the file.
+If your prompts in Langtail contain tools, you can generate a file containing tool parameters for every prompt deployment in your project. Run `npx langtail generate-tools --out [output_filepath]` to generate the file. For typings of the `tools` helper to work correctly, you also need to [generate types](#typed-inputs).
 
 After the file is generated, you can provide the Langtail tools to AI SDK like this:
 ```typescript
@@ -284,10 +287,6 @@ tools(ltModel, {
 })
 ```
 
-## Typed inputs
-
-You can override input types to improve IntelliSense for the `prompt`, `environment`, `version` and `variables` when calling a prompt. Use the command `npx langtail generate-types`.
-
 ## Stream helpers
 
 The AI streams are delivered as JSON objects, which are split into chunks. This can pose a challenge because JSON objects might be distributed across multiple chunks. We have provide you with helper functions to manage these JSON streams more effectively.

package.json

@@ -1,6 +1,6 @@
 {
   "name": "langtail",
-  "version": "0.13.4",
+  "version": "0.13.5",
   "description": "",
   "main": "./Langtail.js",
   "packageManager": "pnpm@8.15.6",

src/bin/langtailTools.ts.template

@@ -1,10 +1,14 @@
 // @ts-ignore
-import { CoreTool } from 'ai'
 import { z } from 'zod'
 
 type Streamable = unknown;
 type Renderer<T extends Array<any>> = (...args: T) => Streamable | Generator<Streamable, Streamable, void> | AsyncGenerator<Streamable, Streamable, void>;
-interface VercelAITool<PARAMETERS extends z.ZodTypeAny = any, RESULT = any> extends CoreTool<PARAMETERS, RESULT> {
+interface VercelAITool<PARAMETERS extends z.ZodTypeAny = any, RESULT = any> {
+  parameters: PARAMETERS;
+  description?: string;
+  execute?: (args: z.infer<PARAMETERS>, options: {
+      abortSignal?: AbortSignal;
+  }) => PromiseLike<RESULT>;
   generate?: Renderer<[
     z.infer<PARAMETERS>,
     {

src/getOpenAIBody.ts

@@ -33,20 +33,23 @@ export function getOpenAIBody(
   parsedBody: IncomingBodyType,
   threadParams?: {
     threadMessages?: PlaygroundMessage[]
-  }
+  },
+  options?: { appendTemplate?: boolean },
 ): ChatCompletionsCreateParams {
   const completionArgs = completionConfig.state.args
 
   const template = parsedBody.template ?? completionConfig.state.template
-  const inputMessages = [
-    ...compileMessages(template, Object.assign(
-      completionConfig.chatInput,
-      parsedBody.variables ?? {},
-    )),
+  const compiledTemplate = compileMessages(template, Object.assign(
+    completionConfig.chatInput,
+    parsedBody.variables ?? {},
+  ))
+  const bodyMessages = [
     ...[...(threadParams?.threadMessages ?? []) as ChatCompletionMessageParam[]],
     ...(parsedBody.messages ?? []) as ChatCompletionMessageParam[]
   ]
 
+  const inputMessages = options?.appendTemplate ? [...bodyMessages, ...compiledTemplate] : [...compiledTemplate, ...bodyMessages]
+
   const openAIbody: OpenAI.Chat.ChatCompletionCreateParams = {
     model: parsedBody.model ?? completionArgs.model,
     temperature: parsedBody.temperature ?? completionArgs.temperature,