kernl 0.1.3 → 0.2.0

package/src/mcp/__tests__/integration.test.ts

@@ -9,9 +9,18 @@ import { tool } from "@/tool";
 import { z } from "zod";
 import { Thread } from "@/thread";
 import { createMCPToolStaticFilter } from "../utils";
+import { createMockModel } from "@/thread/__tests__/fixtures/mock-model";
 
 const TEST_SERVER = path.join(__dirname, "fixtures", "server.ts");
 
+// Mock model for tests that only need toolkit functionality
+const mockModel = createMockModel(async () => ({
+  content: [],
+  finishReason: "stop",
+  usage: { inputTokens: 0, outputTokens: 0, totalTokens: 0 },
+  warnings: [],
+}));
+
 describe("MCP Integration Tests", () => {
   describe("MCPToolkit Integration", () => {
     let server: MCPServerStdio;
@@ -158,6 +167,7 @@ describe("MCP Integration Tests", () => {
         id: "test-agent",
         name: "Test Agent",
         instructions: "Test",
+        model: mockModel,
         toolkits: [toolkit],
       });
 
@@ -194,6 +204,7 @@ describe("MCP Integration Tests", () => {
         id: "test-agent",
         name: "Test Agent",
         instructions: "Test",
+        model: mockModel,
         toolkits: [toolkit],
       });
 
@@ -232,6 +243,7 @@ describe("MCP Integration Tests", () => {
         id: "test-agent",
         name: "Test Agent",
         instructions: "Test",
+        model: mockModel,
         toolkits: [toolkit],
       });
 
@@ -265,6 +277,7 @@ describe("MCP Integration Tests", () => {
         id: "test-agent",
         name: "Test Agent",
         instructions: "Test",
+        model: mockModel,
         toolkits: [mcpToolkit],
       });
 
@@ -294,6 +307,7 @@ describe("MCP Integration Tests", () => {
         id: "test-agent",
         name: "Test Agent",
         instructions: "Test",
+        model: mockModel,
         toolkits: [mcpToolkit],
       });
 
@@ -325,6 +339,7 @@ describe("MCP Integration Tests", () => {
         id: "test-agent",
         name: "Test Agent",
         instructions: "Test",
+        model: mockModel,
         toolkits: [mcpToolkit],
       });
 
@@ -382,6 +397,7 @@ describe("MCP Integration Tests", () => {
         id: "test-agent",
         name: "Test Agent",
         instructions: "Test",
+        model: mockModel,
         toolkits: [mcpToolkit, functionToolkit],
       });
 
@@ -431,6 +447,7 @@ describe("MCP Integration Tests", () => {
         id: "test-agent",
         name: "Test Agent",
         instructions: "Test",
+        model: mockModel,
         toolkits: [mcpToolkit, functionToolkit],
       });
 

package/src/thread/__tests__/fixtures/mock-model.ts

@@ -0,0 +1,71 @@
+import type {
+  LanguageModel,
+  LanguageModelRequest,
+  LanguageModelResponse,
+  LanguageModelItem,
+  LanguageModelStreamEvent,
+} from "@kernl-sdk/protocol";
+
+/**
+ * Helper to convert LanguageModelResponse content to stream events.
+ * Yields both delta events (for streaming UX) and complete items (for history).
+ */
+async function* streamFromResponse(
+  response: LanguageModelResponse,
+): AsyncGenerator<LanguageModelStreamEvent> {
+  for (const item of response.content) {
+    if (item.kind === "message") {
+      // Stream message with text deltas
+      for (const contentItem of item.content) {
+        if (contentItem.kind === "text") {
+          // Yield text-start
+          yield {
+            kind: "text-start" as const,
+            id: item.id,
+          };
+          // Yield text-delta
+          yield {
+            kind: "text-delta" as const,
+            id: item.id,
+            text: contentItem.text,
+          };
+          // Yield text-end
+          yield {
+            kind: "text-end" as const,
+            id: item.id,
+          };
+        }
+      }
+      // Yield complete message
+      yield item;
+    } else {
+      // For tool-call, reasoning, tool-result - just yield as-is
+      yield item;
+    }
+  }
+  // Yield finish event
+  yield {
+    kind: "finish" as const,
+    finishReason: response.finishReason,
+    usage: response.usage,
+  };
+}
+
+/**
+ * Creates a mock LanguageModel that automatically implements streaming
+ * based on the generate() implementation.
+ */
+export function createMockModel(
+  generateFn: (req: LanguageModelRequest) => Promise<LanguageModelResponse>,
+): LanguageModel {
+  return {
+    spec: "1.0" as const,
+    provider: "test",
+    modelId: "test-model",
+    generate: generateFn,
+    stream: async function* (req: LanguageModelRequest) {
+      const response = await generateFn(req);
+      yield* streamFromResponse(response);
+    },
+  };
+}

package/src/thread/__tests__/integration.test.ts

@@ -0,0 +1,349 @@
+import { describe, it, expect, beforeAll } from "vitest";
+import { z } from "zod";
+import { openai } from "@ai-sdk/openai";
+import { AISDKLanguageModel } from "@kernl-sdk/ai";
+
+import { Agent } from "@/agent";
+import { Kernl } from "@/kernl";
+import { tool, Toolkit } from "@/tool";
+
+import { Thread } from "../thread";
+
+import type { ThreadEvent, ThreadStreamEvent } from "@/types/thread";
+
+/**
+ * Integration tests for Thread streaming with real AI SDK providers.
+ *
+ * These tests require an OPENAI_API_KEY environment variable to be set.
+ * They will be skipped if the API key is not available.
+ *
+ * Run with: OPENAI_API_KEY=your-key pnpm test:run
+ */
+
+const SKIP_INTEGRATION_TESTS = !process.env.OPENAI_API_KEY;
+
+describe.skipIf(SKIP_INTEGRATION_TESTS)(
+  "Thread streaming integration",
+  () => {
+    let kernl: Kernl;
+    let model: AISDKLanguageModel;
+
+    beforeAll(() => {
+      kernl = new Kernl();
+      model = new AISDKLanguageModel(openai("gpt-4o"));
+    });
+
+    describe("stream()", () => {
+      it(
+        "should yield both delta events and complete items",
+        async () => {
+      const agent = new Agent({
+        id: "test-stream",
+        name: "Test Stream Agent",
+        instructions: "You are a helpful assistant.",
+        model,
+      });
+
+      const input: ThreadEvent[] = [
+        {
+          kind: "message",
+          id: "msg-1",
+          role: "user",
+          content: [
+            { kind: "text", text: "Say 'Hello World' and nothing else." },
+          ],
+        },
+      ];
+
+      const thread = new Thread(kernl, agent, input);
+      const events = [];
+
+      for await (const event of thread.stream()) {
+        events.push(event);
+      }
+
+      expect(events.length).toBeGreaterThan(0);
+
+      // Should have text-delta events (for streaming UX)
+      const textDeltas = events.filter((e) => e.kind === "text-delta");
+      expect(textDeltas.length).toBeGreaterThan(0);
+
+      // Should have text-start event
+      const textStarts = events.filter((e) => e.kind === "text-start");
+      expect(textStarts.length).toBeGreaterThan(0);
+
+      // Should have text-end event
+      const textEnds = events.filter((e) => e.kind === "text-end");
+      expect(textEnds.length).toBeGreaterThan(0);
+
+      // Should have complete Message item (for history)
+      const messages = events.filter((e) => e.kind === "message");
+      expect(messages.length).toBeGreaterThan(0);
+
+      const assistantMessage = messages.find(
+        (m: any) => m.role === "assistant",
+      );
+      expect(assistantMessage).toBeDefined();
+      expect((assistantMessage as any).content).toBeDefined();
+      expect((assistantMessage as any).content.length).toBeGreaterThan(0);
+
+      // Message should have accumulated text from all deltas
+      const textContent = (assistantMessage as any).content.find(
+        (c: any) => c.kind === "text",
+      );
+      expect(textContent).toBeDefined();
+      expect(textContent.text).toBeDefined();
+      expect(textContent.text.length).toBeGreaterThan(0);
+
+      // Verify accumulated text matches concatenated deltas
+      const accumulatedFromDeltas = textDeltas.map((d: any) => d.text).join("");
+      expect(textContent.text).toBe(accumulatedFromDeltas);
+
+      // Should have finish event
+      const finishEvents = events.filter((e) => e.kind === "finish");
+      expect(finishEvents.length).toBe(1);
+        },
+        30000,
+      );
+
+      it(
+        "should filter deltas from history but include complete items",
+        async () => {
+      const agent = new Agent({
+        id: "test-history",
+        name: "Test History Agent",
+        instructions: "You are a helpful assistant.",
+        model,
+      });
+
+      const input: ThreadEvent[] = [
+        {
+          kind: "message",
+          id: "msg-1",
+          role: "user",
+          content: [{ kind: "text", text: "Count to 3" }],
+        },
+      ];
+
+      const thread = new Thread(kernl, agent, input);
+      const streamEvents = [];
+
+      for await (const event of thread.stream()) {
+        streamEvents.push(event);
+      }
+
+      // Access private history via type assertion for testing
+      const history = (thread as any).history as ThreadEvent[];
+
+      // History should only contain complete items (message, reasoning, tool-call, tool-result)
+      // TypeScript already enforces this via ThreadEvent type, but let's verify at runtime
+      for (const event of history) {
+        expect(["message", "reasoning", "tool-call", "tool-result"]).toContain(
+          event.kind,
+        );
+      }
+
+      // Stream events should include deltas (but history should not)
+      const streamDeltas = streamEvents.filter(
+        (e: any) =>
+          e.kind === "text-delta" ||
+          e.kind === "text-start" ||
+          e.kind === "text-end",
+      );
+      expect(streamDeltas.length).toBeGreaterThan(0);
+
+      // History should contain the input message
+      expect(history[0]).toEqual(input[0]);
+
+      // History should contain complete Message items
+      const historyMessages = history.filter((e) => e.kind === "message");
+      expect(historyMessages.length).toBeGreaterThan(1); // input + assistant response
+
+      // Verify assistant message has complete text (not deltas)
+      const assistantMessage = historyMessages.find(
+        (m: any) => m.role === "assistant",
+      );
+      expect(assistantMessage).toBeDefined();
+      const textContent = (assistantMessage as any).content.find(
+        (c: any) => c.kind === "text",
+      );
+      expect(textContent.text).toBeTruthy();
+      expect(textContent.text.length).toBeGreaterThan(0);
+        },
+        30000,
+      );
+
+      it("should work with tool calls", async () => {
+      const addTool = tool({
+        id: "add",
+        name: "add",
+        description: "Add two numbers together",
+        parameters: z.object({
+          a: z.number().describe("The first number"),
+          b: z.number().describe("The second number"),
+        }),
+        execute: async (ctx, { a, b }) => {
+          return a + b;
+        },
+      });
+
+      const toolkit = new Toolkit({
+        id: "math",
+        tools: [addTool],
+      });
+
+      const agent = new Agent({
+        id: "test-tools",
+        name: "Test Tools Agent",
+        instructions: "You are a helpful assistant that can do math.",
+        model,
+        toolkits: [toolkit],
+      });
+
+      const input: ThreadEvent[] = [
+        {
+          kind: "message",
+          id: "msg-1",
+          role: "user",
+          content: [{ kind: "text", text: "What is 25 + 17?" }],
+        },
+      ];
+
+      const thread = new Thread(kernl, agent, input);
+      const events: ThreadStreamEvent[] = [];
+
+      for await (const event of thread.stream()) {
+        events.push(event);
+      }
+
+      expect(events.length).toBeGreaterThan(0);
+
+      // Should have tool calls
+      const toolCalls = events.filter(
+        (e): e is Extract<ThreadStreamEvent, { kind: "tool-call" }> =>
+          e.kind === "tool-call",
+      );
+      expect(toolCalls.length).toBeGreaterThan(0);
+
+      // Verify tool was called with correct parameters
+      const addToolCall = toolCalls.find((tc) => tc.toolId === "add");
+      expect(addToolCall).toBeDefined();
+      expect(JSON.parse(addToolCall!.arguments)).toEqual({ a: 25, b: 17 });
+
+      // Should have tool results
+      const toolResults = events.filter(
+        (e): e is Extract<ThreadStreamEvent, { kind: "tool-result" }> =>
+          e.kind === "tool-result",
+      );
+      expect(toolResults.length).toBeGreaterThan(0);
+
+      // Verify tool result is correct
+      const addToolResult = toolResults.find(
+        (tr) => tr.callId === addToolCall!.callId,
+      );
+      expect(addToolResult).toBeDefined();
+      expect(addToolResult!.result).toBe(42);
+
+      // History should contain tool calls and results
+      const history = (thread as any).history as ThreadEvent[];
+      const historyToolCalls = history.filter((e) => e.kind === "tool-call");
+      const historyToolResults = history.filter(
+        (e) => e.kind === "tool-result",
+      );
+
+      expect(historyToolCalls.length).toBe(toolCalls.length);
+      expect(historyToolResults.length).toBe(toolResults.length);
+
+      // Verify the assistant's final response references the correct answer
+      const messages = events.filter((e) => e.kind === "message");
+      const assistantMessage = messages.find((m: any) => m.role === "assistant");
+      expect(assistantMessage).toBeDefined();
+      const textContent = (assistantMessage as any).content.find(
+        (c: any) => c.kind === "text",
+      );
+      expect(textContent).toBeDefined();
+      expect(textContent.text).toContain("42");
+        },
+        30000,
+      );
+    });
+
+    describe("execute()", () => {
+      it(
+        "should consume stream and return final response",
+        async () => {
+      const agent = new Agent({
+        id: "test-blocking",
+        name: "Test Blocking Agent",
+        instructions: "You are a helpful assistant.",
+        model,
+      });
+
+      const input: ThreadEvent[] = [
+        {
+          kind: "message",
+          id: "msg-1",
+          role: "user",
+          content: [{ kind: "text", text: "Say 'Testing' and nothing else." }],
+        },
+      ];
+
+      const thread = new Thread(kernl, agent, input);
+      const result = await thread.execute();
+
+      // Should have a response
+      expect(result.response).toBeDefined();
+      expect(typeof result.response).toBe("string");
+      expect(result.response.length).toBeGreaterThan(0);
+
+      // Should have final state
+      expect(result.state).toBe("stopped");
+        },
+        30000,
+      );
+
+      it(
+        "should validate structured output in blocking mode",
+        async () => {
+      const responseSchema = z.object({
+        name: z.string(),
+        age: z.number(),
+      });
+
+      const agent = new Agent({
+        id: "test-structured",
+        name: "Test Structured Agent",
+        instructions:
+          "You are a helpful assistant. Return JSON with name and age fields.",
+        model,
+        responseType: responseSchema,
+      });
+
+      const input: ThreadEvent[] = [
+        {
+          kind: "message",
+          id: "msg-1",
+          role: "user",
+          content: [
+            {
+              kind: "text",
+              text: 'Return a JSON object with name "Alice" and age 30',
+            },
+          ],
+        },
+      ];
+
+      const thread = new Thread(kernl, agent, input);
+      const result = await thread.execute();
+
+      // Response should be validated and parsed
+      expect(result.response).toBeDefined();
+      expect(typeof result.response).toBe("object");
+      expect((result.response as any).name).toBeTruthy();
+      expect(typeof (result.response as any).age).toBe("number");
+        },
+        30000,
+      );
+    });
+  },
+);