feat: make systemPrompt sync and optional (#29)

.changeset/four-spies-mate.md

@@ -0,0 +1,5 @@
+---
+'@callstack/byorg-core': minor
+---
+
+core: make systemPrompt both optional and sync-only

docs/src/docs/core/context.md

@@ -24,7 +24,7 @@ export type RequestContext = {
   resolvedEntities: EntityInfo;
 
   /** Function for generating a system prompt */
-  systemPrompt: () => Promise<string> | string;
+  systemPrompt: () => string | null;
 
   /** 
    * Received partial response update with response streaming.

docs/src/docs/core/usage.mdx

@@ -28,21 +28,13 @@ Here is step by step how to start with byorg.
     languageModel: openAiModel,
   });
   ```
-  ### Prepare a system prompt function
-  ```js
-  import { RequestContext } from '@callstack/byorg-core';
-
-  const systemPrompt = (context: RequestContext): Promise<string> | string => {
-    return "You are a helpful AI assistant named Byorg";
-  };
-  ```
+
   ### Create an Application instance
   ```js
   import { VercelChatModelAdapter } from '@callstack/byorg-core';
 
   const app = createApp({
-    chatModel,
-    systemPrompt,
+    chatModel
   });
   ```
   ### Process user messages

examples/bare/src/index.ts

@@ -19,11 +19,11 @@ const chatModel = new VercelChatModelAdapter({
 // import { createChatMockModel } from '@callstack/byorg-core';
 // const chatModel = createChatMockModel();
 
-const SYSTEM_PROMPT = 'Your name is Byorg. You are an AI Assistant.';
+const SYSTEM_PROMPT = 'Your name is Byorg. You are a helpful AI Assistant.';
 
 const app = createApp({
   chatModel,
-  systemPrompt: () => SYSTEM_PROMPT,
+  systemPrompt: SYSTEM_PROMPT,
 });
 
 // Create a readline interface for user input

examples/discord/src/index.ts

@@ -20,13 +20,11 @@ const chatModel = new VercelChatModelAdapter({
 // import { createChatMockModel } from '@callstack/byorg-core';
 // const chatModel = createChatMockModel();
 
-const systemPrompt = () => {
-  return 'Your name is Byorg. You are a helpful AI Assistant.';
-};
+const SYSTEM_PROMPT = 'Your name is Byorg. You are a helpful AI Assistant.';
 
 const app = createApp({
   chatModel,
-  systemPrompt,
+  systemPrompt: SYSTEM_PROMPT,
 });
 
 const discord = await createDiscordApp({ app });

examples/slack/src/index.ts

@@ -22,11 +22,11 @@ const chatModel = new VercelChatModelAdapter({
 // import { createChatMockModel } from '@callstack/byorg-core';
 // const chatModel = createChatMockModel();
 
-const SYSTEM_PROMPT = 'Your name is Byorg. You are an AI Assistant.';
+const SYSTEM_PROMPT = 'Your name is Byorg. You are a helpful AI Assistant.';
 
 const app = createApp({
   chatModel,
-  systemPrompt: () => SYSTEM_PROMPT,
+  systemPrompt: SYSTEM_PROMPT,
 });
 
 const slack = createSlackApp({

packages/core/src/__tests__/application.test.ts

@@ -9,7 +9,6 @@ test('basic non-streaming test', async () => {
   const testModel = createMockChatModel({ delay: 0, seed: 3 });
   const app = createApp({
     chatModel: testModel,
-    systemPrompt: () => '',
   });
 
   const result = await app.processMessages(messages);
@@ -32,7 +31,6 @@ test('basic streaming test', async () => {
   const testModel = createMockChatModel({ delay: 0, seed: 3 });
   const app = createApp({
     chatModel: testModel,
-    systemPrompt: () => '',
   });
 
   const onPartialResponse = vitest.fn();

packages/core/src/ai/__tests__/vercel.test.ts

@@ -0,0 +1,63 @@
+import { expect, test } from 'vitest';
+import { createMockVercelModel } from '../../mock/vercel-mock-model.js';
+import { createApp } from '../../application.js';
+import { Message } from '../../domain.js';
+import { VercelChatModelAdapter } from '../vercel.js';
+
+const messages: Message[] = [{ role: 'user', content: 'Hello' }];
+
+test('sends system messages when system prompt is provided', async () => {
+  const { languageModel, calls } = createMockVercelModel({ text: 'Hello, world!' });
+  const app = createApp({
+    chatModel: new VercelChatModelAdapter({ languageModel }),
+    systemPrompt: 'This is system prompt',
+  });
+
+  await app.processMessages(messages);
+  expect(calls.length).toBe(1);
+  expect(calls[0].prompt).toContainEqual({
+    role: 'system',
+    content: 'This is system prompt',
+  });
+});
+
+test('does not send system message when system prompt is not provided', async () => {
+  const { languageModel, calls } = createMockVercelModel({ text: 'Hello, world!' });
+  const app = createApp({
+    chatModel: new VercelChatModelAdapter({ languageModel }),
+  });
+
+  await app.processMessages(messages);
+  expect(calls.length).toBe(1);
+  expect(calls[0].prompt).not.toContainEqual({
+    role: 'system',
+  });
+});
+
+test('does not send system message when system prompt returns null', async () => {
+  const { languageModel, calls } = createMockVercelModel({ text: 'Hello, world!' });
+  const app = createApp({
+    chatModel: new VercelChatModelAdapter({ languageModel }),
+    systemPrompt: () => null,
+  });
+
+  await app.processMessages(messages);
+  expect(calls.length).toBe(1);
+  expect(calls[0].prompt).not.toContainEqual({
+    role: 'system',
+  });
+});
+
+test('does not send system message when system prompt returns empty string', async () => {
+  const { languageModel, calls } = createMockVercelModel({ text: 'Hello, world!' });
+  const app = createApp({
+    chatModel: new VercelChatModelAdapter({ languageModel }),
+    systemPrompt: () => '',
+  });
+
+  await app.processMessages(messages);
+  expect(calls.length).toBe(1);
+  expect(calls[0].prompt).not.toContainEqual({
+    role: 'system',
+  });
+});

packages/core/src/ai/vercel.ts

@@ -57,17 +57,20 @@ export class VercelChatModelAdapter implements ChatModel {
   constructor(private readonly _options: VercelChatModelAdapterOptions) {}
 
   async generateResponse(context: RequestContext): Promise<AssistantResponse> {
-    const messages = context.messages;
+    let systemPrompt = context.systemPrompt();
+    if (systemPrompt) {
+      const entitiesPrompt = formatResolvedEntities(context.resolvedEntities);
+      if (entitiesPrompt) {
+        systemPrompt += '\n\n' + entitiesPrompt;
+      }
+    }
 
-    const systemPrompt = await context.systemPrompt();
-    const entitiesPrompt = formatResolvedEntities(context.resolvedEntities);
-    const finalSystemPrompt = [systemPrompt, entitiesPrompt].join('\n\n');
+    const messages: CoreMessage[] = [];
+    if (systemPrompt) {
+      messages.push({ role: 'system', content: systemPrompt });
+    }
 
-    // TODO: Use userId in anonymous case
-    const resolvedMessages: CoreMessage[] = [
-      { role: 'system' as const, content: finalSystemPrompt },
-      ...messages.map(toMessageParam),
-    ];
+    messages.push(...context.messages.map(toMessageParam));
 
     const getRunToolFunction =
       <TParams extends z.ZodSchema, TOutput>(
@@ -94,7 +97,7 @@ export class VercelChatModelAdapter implements ChatModel {
 
     const executionContext: AiExecutionContext = {
       tools,
-      messages: resolvedMessages,
+      messages: messages,
     };
 
     const executionResult = context.onPartialResponse
@@ -229,14 +232,14 @@ function toNumberOrZero(n: number): number {
   return Number.isNaN(n) ? 0 : n;
 }
 
-function formatResolvedEntities(entities: Record<string, unknown>): string {
+function formatResolvedEntities(entities: Record<string, unknown>): string | null {
   if (Object.keys(entities).length === 0) {
-    return '';
+    return null;
   }
 
-  return `ENTITY DICTIONARY: \n
-    ${Object.entries(entities)
-      .map(([key, value]) => `'${key}' is '${JSON.stringify(value)}'`)
-      .join('\n')}
-      `;
+  return `### ENTITY DICTIONARY ###\n
+${Object.entries(entities)
+  .map(([key, value]) => `'${key}' is '${JSON.stringify(value)}'`)
+  .join('\n')}
+  `;
 }

packages/core/src/application.ts

@@ -32,8 +32,8 @@ export type ErrorHandler = (
 ) => Promise<MessageResponse> | MessageResponse;
 
 export type ApplicationConfig = {
-  systemPrompt: (context: RequestContext) => Promise<string> | string;
   chatModel: ChatModel | ((context: RequestContext) => ChatModel);
+  systemPrompt?: ((context: RequestContext) => string | null) | string;
   plugins?: ApplicationPlugin[];
   errorHandler?: ErrorHandler;
 };
@@ -89,8 +89,6 @@ export function createApp(config: ApplicationConfig): Application {
       const performance = new PerformanceTimeline();
       performance.markStart(PerformanceMarks.processMessages);
 
-      const onPartialResponse = options?.onPartialResponse;
-
       const context: RequestContext = {
         messages,
         get lastMessage() {
@@ -102,14 +100,16 @@ export function createApp(config: ApplicationConfig): Application {
 
           return lastMessage;
         },
-
+        systemPrompt: () =>
+          typeof config.systemPrompt === 'function'
+            ? config.systemPrompt(context)
+            : (config.systemPrompt ?? null),
+        onPartialResponse: options?.onPartialResponse,
         tools,
         references: getReferenceStorage(),
         resolvedEntities: {},
-        onPartialResponse,
         extras: options?.extras ?? {},
         performance,
-        systemPrompt: () => config.systemPrompt(context),
       };
 
       const handler = async () => {

packages/core/src/domain.ts

@@ -30,7 +30,7 @@ export type RequestContext = {
   tools: ApplicationTool[];
   references: ReferenceStorage;
   resolvedEntities: EntityInfo;
-  systemPrompt: () => Promise<string> | string;
+  systemPrompt: () => string | null;
   onPartialResponse?: (text: string) => void;
   extras: MessageRequestExtras;
   performance: PerformanceTimeline;

packages/core/src/mock/vercel-mock-model.ts

@@ -0,0 +1,24 @@
+import { LanguageModelV1CallOptions } from 'ai';
+import { MockLanguageModelV1 } from 'ai/test';
+
+export type MockVercelModelOptions = {
+  text: string;
+};
+
+export function createMockVercelModel({ text }: MockVercelModelOptions) {
+  const calls: LanguageModelV1CallOptions[] = [];
+  const languageModel = new MockLanguageModelV1({
+    doGenerate: (options) => {
+      calls.push(options);
+      return Promise.resolve({
+        rawCall: { rawPrompt: null, rawSettings: {} },
+        finishReason: 'stop',
+        usage: { promptTokens: 10, completionTokens: 20 },
+        text,
+      });
+    },
+    // TODO: add doStream when needed
+  });
+
+  return { languageModel, calls };
+}