diff --git a/packages/core/src/services/ai/index.ts b/packages/core/src/services/ai/index.ts
index 1388ef470..319fd6f0c 100644
--- a/packages/core/src/services/ai/index.ts
+++ b/packages/core/src/services/ai/index.ts
@@ -50,8 +50,8 @@ export async function ai({
   documentLogUuid,
   source,
   onFinish,
-  schema = config.schema,
-  output = config.schema?.type || 'no-schema',
+  schema,
+  output,
   transactionalLogs = false,
 }: {
   workspace: Workspace
diff --git a/packages/core/src/services/chains/run.test.ts b/packages/core/src/services/chains/run.test.ts
index 2ede8bfc4..c22de6a25 100644
--- a/packages/core/src/services/chains/run.test.ts
+++ b/packages/core/src/services/chains/run.test.ts
@@ -87,7 +87,7 @@ describe('runChain', () => {
       expect.objectContaining({
         config: { provider: 'openai', model: 'gpt-3.5-turbo' },
         schema: undefined,
-        output: undefined,
+        output: 'no-schema',
       }),
     )
   })
@@ -299,4 +299,200 @@ describe('runChain', () => {
       }),
     )
   })
+
+  it('runs a chain with object schema and output', async () => {
+    const mockSchema = {
+      type: 'object',
+      properties: {
+        name: { type: 'string' },
+        age: { type: 'number' },
+      },
+    } as const
+
+    const mockAiResponse = {
+      object: Promise.resolve({ name: 'John', age: 30 }),
+      usage: Promise.resolve({ totalTokens: 15 }),
+      fullStream: new ReadableStream({
+        start(controller) {
+          controller.enqueue({
+            type: 'object',
+            object: { name: 'John', age: 30 },
+          })
+          controller.close()
+        },
+      }),
+    }
+
+    vi.spyOn(aiModule, 'ai').mockResolvedValue(mockAiResponse as any)
+
+    vi.mocked(mockChain.step!).mockResolvedValue({
+      completed: true,
+      conversation: {
+        messages: [
+          {
+            role: MessageRole.user,
+            content: [{ type: ContentType.text, text: 'Test message' }],
+          },
+        ],
+        config: { provider: 'openai', model: 'gpt-3.5-turbo' },
+      },
+    })
+
+    const result = await runChain({
+      workspace,
+      chain: mockChain as Chain,
+      apikeys,
+      source: LogSources.API,
+      configOverrides: {
+        schema: mockSchema,
+        output: 'object',
+      },
+    })
+
+    expect(result.ok).toBe(true)
+    if (!result.ok) return
+
+    const response = await result.value.response
+    expect(response).toEqual({
+      documentLogUuid: expect.any(String),
+      object: { name: 'John', age: 30 },
+      text: '{"name":"John","age":30}',
+      usage: { totalTokens: 15 },
+    })
+
+    expect(aiModule.ai).toHaveBeenCalledWith(
+      expect.objectContaining({
+        schema: mockSchema,
+        output: 'object',
+      }),
+    )
+  })
+
+  it('runs a chain with array schema and output', async () => {
+    const mockSchema = {
+      type: 'array',
+      items: {
+        type: 'object',
+        properties: {
+          name: { type: 'string' },
+          age: { type: 'number' },
+        },
+      },
+    } as const
+
+    const mockAiResponse = {
+      object: Promise.resolve([
+        { name: 'John', age: 30 },
+        { name: 'Jane', age: 25 },
+      ]),
+      usage: Promise.resolve({ totalTokens: 20 }),
+      fullStream: new ReadableStream({
+        start(controller) {
+          controller.enqueue({
+            type: 'object',
+            object: [
+              { name: 'John', age: 30 },
+              { name: 'Jane', age: 25 },
+            ],
+          })
+          controller.close()
+        },
+      }),
+    }
+
+    vi.spyOn(aiModule, 'ai').mockResolvedValue(mockAiResponse as any)
+
+    vi.mocked(mockChain.step!).mockResolvedValue({
+      completed: true,
+      conversation: {
+        messages: [
+          {
+            role: MessageRole.user,
+            content: [{ type: ContentType.text, text: 'Test message' }],
+          },
+        ],
+        config: { provider: 'openai', model: 'gpt-3.5-turbo' },
+      },
+    })
+
+    const result = await runChain({
+      workspace,
+      chain: mockChain as Chain,
+      apikeys,
+      source: LogSources.API,
+      configOverrides: {
+        schema: mockSchema,
+        output: 'array',
+      },
+    })
+
+    expect(result.ok).toBe(true)
+    if (!result.ok) return
+
+    const response = await result.value.response
+    expect(response).toEqual({
+      documentLogUuid: expect.any(String),
+      object: [
+        { name: 'John', age: 30 },
+        { name: 'Jane', age: 25 },
+      ],
+      text: '[{"name":"John","age":30},{"name":"Jane","age":25}]',
+      usage: { totalTokens: 20 },
+    })
+
+    expect(aiModule.ai).toHaveBeenCalledWith(
+      expect.objectContaining({
+        schema: mockSchema,
+        output: 'array',
+      }),
+    )
+  })
+
+  it('runs a chain with no-schema output', async () => {
+    const mockAiResponse = createMockAiResponse(
+      'AI response without schema',
+      10,
+    )
+    vi.spyOn(aiModule, 'ai').mockResolvedValue(mockAiResponse as any)
+
+    vi.mocked(mockChain.step!).mockResolvedValue({
+      completed: true,
+      conversation: {
+        messages: [
+          {
+            role: MessageRole.user,
+            content: [{ type: ContentType.text, text: 'Test message' }],
+          },
+        ],
+        config: { provider: 'openai', model: 'gpt-3.5-turbo' },
+      },
+    })
+
+    const result = await runChain({
+      workspace,
+      chain: mockChain as Chain,
+      apikeys,
+      source: LogSources.API,
+      configOverrides: {
+        output: 'no-schema',
+      },
+    })
+
+    expect(result.ok).toBe(true)
+    if (!result.ok) return
+
+    const response = await result.value.response
+    expect(response).toEqual({
+      documentLogUuid: expect.any(String),
+      text: 'AI response without schema',
+      usage: { totalTokens: 10 },
+      toolCalls: [],
+    })
+
+    expect(aiModule.ai).toHaveBeenCalledWith(
+      expect.objectContaining({
+        output: 'no-schema',
+      }),
+    )
+  })
 })
diff --git a/packages/core/src/services/chains/run.ts b/packages/core/src/services/chains/run.ts
index 3ec3a9fa0..ff9577438 100644
--- a/packages/core/src/services/chains/run.ts
+++ b/packages/core/src/services/chains/run.ts
@@ -20,6 +20,12 @@ import { streamToGenerator } from '../../lib/streamToGenerator'
 import { ai, Config, validateConfig } from '../ai'
 
 export type CachedApiKeys = Map<string, ProviderApiKey>
+type ConfigOverrides =
+  | {
+      schema: JSONSchema7
+      output: 'object' | 'array'
+    }
+  | { output: 'no-schema' }
 
 export async function runChain({
   workspace,
@@ -34,10 +40,7 @@ export async function runChain({
   generateUUID?: () => string
   source: LogSources
   apikeys: CachedApiKeys
-  configOverrides?: {
-    schema: JSONSchema7
-    output: 'object' | 'array' | 'no-schema'
-  }
+  configOverrides?: ConfigOverrides
 }) {
   const documentLogUuid = generateUUID()
 
@@ -96,13 +99,10 @@ async function iterate({
   previousApiKey?: ProviderApiKey
   documentLogUuid: string
   previousResponse?: ChainTextResponse
-  configOverrides?: {
-    schema: JSONSchema7
-    output: 'object' | 'array' | 'no-schema'
-  }
+  configOverrides?: ConfigOverrides
 }) {
   try {
-    const stepResult = await computeStepData({
+    const step = await computeStepData({
       chain,
       previousResponse,
       apikeys,
@@ -110,26 +110,26 @@ async function iterate({
       sentCount: previousCount,
     })
 
-    publishStepStartEvent(controller, stepResult)
+    publishStepStartEvent(controller, step)
 
     const aiResult = await ai({
       workspace,
       source,
       documentLogUuid,
-      messages: stepResult.conversation.messages,
-      config: stepResult.config,
-      provider: stepResult.apiKey,
-      schema: configOverrides?.schema,
-      output: configOverrides?.output,
-      transactionalLogs: stepResult.completed,
+      messages: step.conversation.messages,
+      config: step.config,
+      provider: step.apiKey,
+      schema: getSchemaForAI(step, configOverrides),
+      output: getOutputForAI(step, configOverrides),
+      transactionalLogs: step.completed,
     })
 
     await streamAIResult(controller, aiResult)
 
     const response = await createChainResponse(aiResult, documentLogUuid)
 
-    if (stepResult.completed) {
-      await handleCompletedChain(controller, stepResult, response)
+    if (step.completed) {
+      await handleCompletedChain(controller, step, response)
       return response
     } else {
       publishStepCompleteEvent(controller, response)
@@ -141,8 +141,8 @@ async function iterate({
         documentLogUuid,
         apikeys,
         controller,
-        previousApiKey: stepResult.apiKey,
-        previousCount: stepResult.sentCount,
+        previousApiKey: step.apiKey,
+        previousCount: step.sentCount,
         previousResponse: response as ChainTextResponse,
         configOverrides,
       })
@@ -153,7 +153,24 @@ async function iterate({
   }
 }
 
-// Helper functions
+function getSchemaForAI(
+  step: Awaited<ReturnType<typeof computeStepData>>,
+  configOverrides?: ConfigOverrides,
+) {
+  return step.completed
+    ? // @ts-expect-error - schema does not exist in some types of configOverrides which is fine
+      configOverrides?.schema || step.config.schema
+    : undefined
+}
+
+function getOutputForAI(
+  step: Awaited<ReturnType<typeof computeStepData>>,
+  configOverrides?: ConfigOverrides,
+) {
+  return step.completed
+    ? configOverrides?.output || step.config.schema?.type || 'no-schema'
+    : undefined
+}
 
 function publishStepStartEvent(
   controller: ReadableStreamDefaultController,