diff --git a/src/core/langfuse/LangfusePromptGateway.ts b/src/core/langfuse/LangfusePromptGateway.ts
new file mode 100644
index 0000000..9950348
--- /dev/null
+++ b/src/core/langfuse/LangfusePromptGateway.ts
@@ -0,0 +1,25 @@
+import { Langfuse } from 'langfuse';
+import type { PromptFetcher, RawPrompt } from './PromptGateway';
+
+export class LangfusePromptGateway implements PromptFetcher {
+  private readonly client: Langfuse;
+
+  constructor(opts: { publicKey: string; secretKey: string; baseUrl: string }) {
+    this.client = new Langfuse({
+      publicKey: opts.publicKey,
+      secretKey: opts.secretKey,
+      baseUrl: opts.baseUrl,
+    });
+  }
+
+  async fetch(promptName: string): Promise<RawPrompt> {
+    const prompt = await this.client.getPrompt(promptName, undefined, { type: 'chat', cacheTtlSeconds: 0 });
+    return {
+      name: prompt.name,
+      version: prompt.version,
+      config: (prompt.config ?? {}) as Record<string, unknown>,
+      compile: (variables) =>
+        prompt.compile(variables) as { role: 'system' | 'user' | 'assistant'; content: string }[],
+    };
+  }
+}
diff --git a/src/core/langfuse/LangfuseTracer.ts b/src/core/langfuse/LangfuseTracer.ts
new file mode 100644
index 0000000..8a41bb2
--- /dev/null
+++ b/src/core/langfuse/LangfuseTracer.ts
@@ -0,0 +1,40 @@
+import { Langfuse } from 'langfuse';
+import { Tracer } from '../llm/Tracer';
+import type { GenerationHandle, GenerationInput, GenerationOutput, TraceHandle } from '../llm/Tracer';
+
+export class LangfuseTracer extends Tracer {
+  private readonly client: Langfuse;
+
+  constructor(opts: { publicKey: string; secretKey: string; baseUrl: string }) {
+    super();
+    this.client = new Langfuse({
+      publicKey: opts.publicKey,
+      secretKey: opts.secretKey,
+      baseUrl: opts.baseUrl,
+    });
+  }
+
+  startTrace(name: string, metadata?: Record<string, unknown>): TraceHandle {
+    const trace = this.client.trace({ name, metadata });
+    return {
+      id: trace.id,
+      startGeneration(input: GenerationInput): GenerationHandle {
+        const generation = trace.generation({
+          name: input.name,
+          model: input.model,
+          input: input.input,
+          metadata: input.metadata,
+        });
+        return {
+          end(output: GenerationOutput) {
+            generation.end({ output: output.output, usage: output.usage, metadata: output.metadata });
+          },
+        };
+      },
+    };
+  }
+
+  async flush(): Promise<void> {
+    await this.client.flushAsync();
+  }
+}
diff --git a/test/core/langfuse/LangfusePromptGateway.test.ts b/test/core/langfuse/LangfusePromptGateway.test.ts
new file mode 100644
index 0000000..48f2d71
--- /dev/null
+++ b/test/core/langfuse/LangfusePromptGateway.test.ts
@@ -0,0 +1,23 @@
+import { describe, expect, it, vi } from 'vitest';
+
+const compile = vi.fn().mockReturnValue([{ role: 'user', content: 'hi' }]);
+const getPrompt = vi.fn().mockResolvedValue({ name: 'greet', version: 2, config: { model: 'llama3' }, compile });
+
+vi.mock('langfuse', () => ({
+  Langfuse: vi.fn().mockImplementation(() => ({ getPrompt })),
+}));
+
+import { LangfusePromptGateway } from '../../../src/core/langfuse/LangfusePromptGateway';
+
+describe('LangfusePromptGateway', () => {
+  it('fetches a chat prompt from Langfuse bypassing the SDK cache, and adapts it to RawPrompt', async () => {
+    const gateway = new LangfusePromptGateway({ publicKey: 'pk', secretKey: 'sk', baseUrl: 'http://lf' });
+    const raw = await gateway.fetch('greet');
+
+    expect(getPrompt).toHaveBeenCalledWith('greet', undefined, { type: 'chat', cacheTtlSeconds: 0 });
+    expect(raw.name).toBe('greet');
+    expect(raw.version).toBe(2);
+    expect(raw.config).toEqual({ model: 'llama3' });
+    expect(raw.compile({ who: 'alice' })).toEqual([{ role: 'user', content: 'hi' }]);
+  });
+});
diff --git a/test/core/langfuse/LangfuseTracer.test.ts b/test/core/langfuse/LangfuseTracer.test.ts
new file mode 100644
index 0000000..2da0af9
--- /dev/null
+++ b/test/core/langfuse/LangfuseTracer.test.ts
@@ -0,0 +1,42 @@
+import { describe, expect, it, vi } from 'vitest';
+
+const generationEnd = vi.fn();
+const generation = vi.fn().mockReturnValue({ end: generationEnd });
+const trace = vi.fn().mockReturnValue({ id: 'lf-trace-1', generation });
+const flushAsync = vi.fn().mockResolvedValue(undefined);
+
+vi.mock('langfuse', () => ({
+  Langfuse: vi.fn().mockImplementation(() => ({ trace, flushAsync })),
+}));
+
+import { LangfuseTracer } from '../../../src/core/langfuse/LangfuseTracer';
+
+describe('LangfuseTracer', () => {
+  it('starts a Langfuse trace and exposes its id', () => {
+    const tracer = new LangfuseTracer({ publicKey: 'pk', secretKey: 'sk', baseUrl: 'http://lf' });
+    const handle = tracer.startTrace('my-pipeline', { foo: 'bar' });
+
+    expect(trace).toHaveBeenCalledWith({ name: 'my-pipeline', metadata: { foo: 'bar' } });
+    expect(handle.id).toBe('lf-trace-1');
+  });
+
+  it('startGeneration creates a Langfuse generation and end() forwards output/usage', () => {
+    const tracer = new LangfuseTracer({ publicKey: 'pk', secretKey: 'sk', baseUrl: 'http://lf' });
+    const handle = tracer.startTrace('my-pipeline');
+    const gen = handle.startGeneration({ name: 'step-1', model: 'llama3', input: ['hi'] });
+    gen.end({ output: 'done', usage: { totalTokens: 5 } });
+
+    expect(generation).toHaveBeenCalledWith(
+      expect.objectContaining({ name: 'step-1', model: 'llama3', input: ['hi'] })
+    );
+    expect(generationEnd).toHaveBeenCalledWith(
+      expect.objectContaining({ output: 'done', usage: { totalTokens: 5 } })
+    );
+  });
+
+  it('flush() calls flushAsync on the underlying client', async () => {
+    const tracer = new LangfuseTracer({ publicKey: 'pk', secretKey: 'sk', baseUrl: 'http://lf' });
+    await tracer.flush();
+    expect(flushAsync).toHaveBeenCalledTimes(1);
+  });
+});