### Example Project Setup

Source: https://github.com/restatedev/ai-examples/blob/main/vercel-ai/tour-of-agents/README.md

Commands to get the example project and install its dependencies.

```bash
restate example typescript-vercel-ai-tour-of-agents && cd typescript-vercel-ai-tour-of-agents
npm install
```

--------------------------------

### Run example

Source: https://github.com/restatedev/ai-examples/blob/main/google-adk/examples/README.md

To get started, cd into any of the examples and then run them.

```shell
uv run .
```

--------------------------------

### Starting the Agents NextJS app

Source: https://github.com/restatedev/ai-examples/blob/main/vercel-ai/nextjs-example-app/README.md

Commands to install dependencies and run the NextJS development server.

```bash
npm install
npm run dev
```

--------------------------------

### Install Restate

Source: https://github.com/restatedev/ai-examples/blob/main/openai-agents/examples/langfuse/README.md

Install Restate via brew or other installation methods.

```bash
brew install restatedev/tap/restate-server restatedev/tap/restate
```

--------------------------------

### Starting Restate Server

Source: https://github.com/restatedev/ai-examples/blob/main/vercel-ai/nextjs-example-app/README.md

Command to start the Restate server.

```shell
npx @restatedev/restate-server@latest
```

--------------------------------

### Start the agent service

Source: https://github.com/restatedev/ai-examples/blob/main/openai-agents/examples/langfuse/README.md

Start the agent service using uvicorn.

```bash
uv run --env-file .env .
```

--------------------------------

### Start NextJS App

Source: https://github.com/restatedev/ai-examples/blob/main/vercel-ai/nextjs-template/README.md

Starts the NextJS application which includes the agent code.

```shell
npm run dev
```

--------------------------------

### Download the example

Source: https://github.com/restatedev/ai-examples/blob/main/google-adk/examples/langfuse/README.md

Commands to download the example project and navigate to the Langfuse directory.

```bash
restate example python-google-adk-examples
cd python-google-adk-examples/langfuse
```

--------------------------------

### Download the example

Source: https://github.com/restatedev/ai-examples/blob/main/openai-agents/examples/langfuse/README.md

Download the example code for the Langfuse integration.

```bash
restate example python-openai-agents-examples
cd python-openai-agents-examples/langfuse
```

--------------------------------

### Download Example

Source: https://github.com/restatedev/ai-examples/blob/main/openai-agents/examples/arize_phoenix/README.md

Commands to download the example project and navigate into the directory.

```bash
restate example python-openai-agents-examples
cd python-openai-agents-examples/arize_phoenix
```

--------------------------------

### Download the example

Source: https://github.com/restatedev/ai-examples/blob/main/python-restate-only/examples/langfuse/README.md

Downloads the example code for the Restate agent and Langfuse integration.

```bash
restate example python-restate-agent-examples
cd python-restate-agent-examples/langfuse
```

--------------------------------

### Get the example

Source: https://github.com/restatedev/ai-examples/blob/main/typescript-restate-only/template/README.md

Clones the TypeScript Restate agent example and installs its dependencies.

```bash
restate example typescript-restate-agent && cd typescript-restate-agent
npm install
```

--------------------------------

### Download the example

Source: https://github.com/restatedev/ai-examples/blob/main/pydantic-ai/examples/langfuse/README.md

Downloads the Python Pydantic AI Langfuse example.

```bash
restate example python-pydantic-ai-examples
cd python-pydantic-ai-examples/langfuse
```

--------------------------------

### Run example

Source: https://github.com/restatedev/ai-examples/blob/main/openai-agents/examples/README.md

Command to run an example after changing directory into it.

```shell
uv run .

```

--------------------------------

### Start Tools Services

Source: https://github.com/restatedev/ai-examples/blob/main/mcp/README.md

Commands to navigate to the tools directory, install dependencies, and run the app.

```shell
cd tools
npm install
npm run app

```

--------------------------------

### Start Restate server and export Langfuse API keys

Source: https://github.com/restatedev/ai-examples/blob/main/openai-agents/examples/langfuse/README.md

Start the Restate server and configure it to export traces to Langfuse by setting the tracing endpoint and headers.

```bash
# Export Langfuse API keys
source .env 
export RESTATE_TRACING_HEADERS__AUTHORIZATION="Basic $(echo -n "${LANGFUSE_PUBLIC_KEY}:${LANGFUSE_SECRET_KEY}" | base64)"

restate-server --tracing-endpoint otlp+https://cloud.langfuse.com/api/public/otel/v1/traces
```

--------------------------------

### Download the example

Source: https://github.com/restatedev/ai-examples/blob/main/pydantic-ai/examples/logfire/README.md

Downloads the Python Pydantic AI examples and navigates into the logfire directory.

```bash
restate example python-pydantic-ai-examples
cd python-pydantic-ai-examples/logfire
```

--------------------------------

### Start Restate Server with Tracing

Source: https://github.com/restatedev/ai-examples/blob/main/openai-agents/examples/arize_phoenix/README.md

Command to start the Restate server and configure it to export OpenTelemetry traces to Arize Phoenix.

```bash
source .env
export RESTATE_TRACING_HEADERS__AUTHORIZATION="Bearer ${PHOENIX_API_KEY}"

restate-server --tracing-endpoint otlp+https://app.phoenix.arize.com/s/your-account-name/v1/traces
```

--------------------------------

### Real-time Updates via PubSub Example

Source: https://github.com/restatedev/ai-examples/blob/main/vercel-ai/nextjs-example-app/README.md

Example of an agent publishing intermediate results to a pubsub topic.

```typescript
onStepFinish: async (step) => {
  step.toolCalls.forEach((toolCall) => {
    pubsub.publish(topic, {
      role: "assistant",
      content: `Tool call: ${toolCall.toolName}(${JSON.stringify(toolCall.input)})`
    }, ctx.rand.uuidv4());
  });
},
```

--------------------------------

### Add API keys to .env file

Source: https://github.com/restatedev/ai-examples/blob/main/google-adk/examples/langfuse/README.md

Example of how to set up the .env file with necessary API keys for Langfuse and Google.

```bash
echo 'LANGFUSE_PUBLIC_KEY=pk-lf-...' > .env                                                                                                                                   
echo 'LANGFUSE_SECRET_KEY=sk-lf-...' >> .env              
echo 'LANGFUSE_BASE_URL=https://cloud.langfuse.com' >> .env 
echo 'GOOGLE_API_KEY=AI...' >> .env 
```

--------------------------------

### Durable Tool Execution Example

Source: https://github.com/restatedev/ai-examples/blob/main/vercel-ai/nextjs-example-app/README.md

Example of wrapping tool calls into durable steps using ctx.run.

```typescript
execute: async ({ expression }) => {
  return await ctx.run(
    `evaluating ${expression}`,
    async () => mathjs.evaluate(expression),
    { serde: superJson }
  );
},
```

--------------------------------

### Install Dependencies

Source: https://github.com/restatedev/ai-examples/blob/main/typescript-restate-only/examples/interrupt-regenerate/README.md

Installs project dependencies using npm.

```bash
npm install
```

--------------------------------

### Start Restate with Logfire tracing

Source: https://github.com/restatedev/ai-examples/blob/main/pydantic-ai/examples/logfire/README.md

Starts the Restate server and configures it to export OpenTelemetry traces to Logfire.

```bash
# Export Logfire write token (=! API key)
source .env 
export RESTATE_TRACING_HEADERS__AUTHORIZATION="Bearer $LOGFIRE_WRITE_TOKEN"

restate-server --tracing-endpoint otlp+https://logfire-eu.pydantic.dev/v1/traces
```

--------------------------------

### Install Restate Server & CLI

Source: https://github.com/restatedev/ai-examples/blob/main/typescript-restate-only/examples/interrupt-regenerate/README.md

Installs the Restate server and CLI using npm.

```bash
npx @restatedev/restate-server
```

--------------------------------

### Start Restate

Source: https://github.com/restatedev/ai-examples/blob/main/google-adk/tour-of-agents/README.md

Start Restate using Docker.

```bash
docker run --name restate_dev --rm \
  -p 8080:8080 -p 9070:9070 -p 9071:9071 \
  --add-host=host.docker.internal:host-gateway \
  docker.restate.dev/restatedev/restate:latest
```

--------------------------------

### Running a Specific Agent

Source: https://github.com/restatedev/ai-examples/blob/main/vercel-ai/tour-of-agents/README.md

Example command to export the OpenAI API key and run the chat-agent.ts example.

```bash
export OPENAI_API_KEY=sk-...
npx tsx ./src/chat-agent.ts
```

--------------------------------

### Start Restate Server

Source: https://github.com/restatedev/ai-examples/blob/main/mcp/README.md

Command to start the Restate Server.

```shell
restate-server

```

--------------------------------

### Durable LLM Calls Example

Source: https://github.com/restatedev/ai-examples/blob/main/vercel-ai/nextjs-example-app/README.md

Example of wrapping an LLM model for durable journaling of inference steps.

```typescript
const model = wrapLanguageModel({
  model: openai("gpt-5.4"),
  middleware: durableCalls(ctx, { maxRetryAttempts: 3 }),
});
```

--------------------------------

### Get History (Interruption Path)

Source: https://github.com/restatedev/ai-examples/blob/main/python-restate-only/examples/interrupt-regenerate/README.md

Curl command to get the conversation history after an interruption.

```bash
curl localhost:8080/CodingAgent/bob/get_history
```

--------------------------------

### Get History (Happy Path)

Source: https://github.com/restatedev/ai-examples/blob/main/python-restate-only/examples/interrupt-regenerate/README.md

Curl command to get the conversation history after a successful task.

```bash
curl localhost:8080/CodingAgent/alice/get_history
```

--------------------------------

### Start the Agent

Source: https://github.com/restatedev/ai-examples/blob/main/typescript-restate-only/tour-of-agents/README.md

Start the agent using tsx.

```shell
npx tsx ./src/chat-agent.ts
```

--------------------------------

### Invoking the agent via HTTP

Source: https://github.com/restatedev/ai-examples/blob/main/vercel-ai/nextjs-example-app/README.md

Example curl command to invoke the agent directly via HTTP.

```shell
curl localhost:8080/agent/chat --json '{ "prompt": "A taxi driver earns $9461 per 1-hour of work. If he works 12 hours a day and in 1 hour he uses 12 liters of petrol with a price of $134 for 1 liter. How much money does he earn in one day?", "topic": "my-session" }'
```

--------------------------------

### Start the Restate Server

Source: https://github.com/restatedev/ai-examples/blob/main/typescript-restate-only/tour-of-agents/README.md

Start the Restate Server in a separate shell.

```shell
npm install --global @restatedev/restate-server@latest @restatedev/restate@latest
restate-server
```

--------------------------------

### Install Restate

Source: https://github.com/restatedev/ai-examples/blob/main/typescript-restate-only/examples/braintrust/README.md

Install Restate server and client globally.

```bash
npm install --global @restatedev/restate-server@latest @restatedev/restate@latest
```

--------------------------------

### Start Restate Server

Source: https://github.com/restatedev/ai-examples/blob/main/google-adk/template/README.md

Start the Restate Server in a separate shell.

```shell
restate-server
```

--------------------------------

### Google API Key Setup

Source: https://github.com/restatedev/ai-examples/blob/main/a2a/README.md

This snippet shows how to set up the Google API Key by adding it to a .env file.

```shell
echo "GOOGLE_API_KEY=your_api_key_here" >> .env
```

--------------------------------

### Add API keys to .env file

Source: https://github.com/restatedev/ai-examples/blob/main/openai-agents/examples/langfuse/README.md

Add your Langfuse and OpenAI API keys to a .env file.

```bash
echo 'LANGFUSE_PUBLIC_KEY=pk-lf-...' > .env                                                                                                                                   
echo 'LANGFUSE_SECRET_KEY=sk-lf-...' >> .env              
echo 'LANGFUSE_BASE_URL=https://cloud.langfuse.com' >> .env 
echo 'OPENAI_API_KEY=sk-proj-...' >> .env 
```

--------------------------------

### Start Agent Service

Source: https://github.com/restatedev/ai-examples/blob/main/typescript-restate-only/examples/interrupt-regenerate/README.md

Starts the agent service with an OpenAI API key.

```bash
OPENAI_API_KEY=sk-proj-... npm run dev
```

--------------------------------

### Set OpenAI API Key

Source: https://github.com/restatedev/ai-examples/blob/main/python-restate-only/examples/interrupt-regenerate/README.md

Command to set the OpenAI API key in an .env file.

```bash
echo 'OPENAI_API_KEY=sk-proj-...' > .env
```

--------------------------------

### Durable Service Handler Example

Source: https://github.com/restatedev/ai-examples/blob/main/vercel-ai/nextjs-example-app/README.md

Example of an AI agent function running as a Restate durable service handler.

```typescript
export default restate.service({
  name: "agent",
  handlers: {
    chat: restate.handlers.handler(
      {
        /* schema */
      },
      async (ctx: restate.Context, { prompt, topic }) => {
        return await runAgent(ctx, prompt, topic);
      },
    ),
  },
});
```

--------------------------------

### Restate-LangChain Integration

Source: https://github.com/restatedev/ai-examples/blob/main/langchain-python/tour-of-agents/README.md

Example of how to integrate RestateMiddleware with LangChain's create_agent.

```python
from langchain.agents import create_agent
from restate.ext.langchain import RestateMiddleware

agent = create_agent(model=..., tools=[...], middleware=[RestateMiddleware()])
```

--------------------------------

### Register the service with Restate

Source: https://github.com/restatedev/ai-examples/blob/main/openai-agents/examples/langfuse/README.md

Register the agent service with Restate so it knows where it is running.

```bash
restate deployments add localhost:9080
```

--------------------------------

### Send a request to the agent

Source: https://github.com/restatedev/ai-examples/blob/main/google-adk/examples/langfuse/README.md

Example cURL command to send a request to the insurance claim agent service running on Restate.

```bash
curl localhost:8080/InsuranceClaimAgent/user123/run \
  --json '{"text": "Customer ID: cus_123 - Hospital bill for broken leg treatment at General Hospital for 3000 euro on 24/04/26"}'
```

--------------------------------

### Task Output Example

Source: https://github.com/restatedev/ai-examples/blob/main/a2a/README.md

This is an example of the JSON output received when querying for task details.

```json
{
  "id": 2,
  "jsonrpc": "2.0",
  "result": {
    "artifacts": [
      {
        "artifactId": "d2ab957f-003d-49f7-b96d-1a5a83fee3b9",
        "description": null,
        "extensions": null,
        "metadata": null,
        "name": null,
        "parts": [
          {
            "kind": "text",
            "metadata": null,
            "text": "Your reimbursement request with ID da52bd93-1927-4a62-8a15-eaafb29dd899 has been approved."
          }
        ]
      }
    ],
    "contextId": "5e00b4a6-dcc67-43ee-a389-0e2a65958444",
    "history": [
      {
        "contextId": "5e00b4a6-dcc67-43ee-a389-0e2a65958444",
        "extensions": null,
        "kind": "message",
        "messageId": "9224947643702-7674c-417b-a0b0-f0741243c589",
        "metadata": null,
        "parts": [
          {
            "kind": "text",
            "metadata": null,
            "text": "Reimburse my hotel for my business trip of 5 nights for 1200USD"
          }
        ],
        "referenceTaskIds": null,
        "role": "user",
        "taskId": "0d9g9ea6-dcc67-43ee-a389-8s9g0e6es5554"
      },
      {
        "contextId": "5e00b4a6-dcc67-43ee-a389-0e2a65958444",
        "extensions": null,
        "kind": "message",
        "messageId": "92249e73702-767c-417b-a06b0-5sg6g881243c589",
        "metadata": null,
        "parts": [
          {
            "kind": "text",
            "metadata": null,
            "text": "The date of the transaction is 05/04/2025"
          }
        ],
        "referenceTaskIds": null,
        "role": "user",
        "taskId": "0d9g9ea6-dcc67-43ee-a389-8s9g0e6es5554"
      },
      {
        "contextId": null,
        "extensions": null,
        "kind": "message",
        "messageId": "89194646-702b-4af3-86d5-7c60355fbb77",
        "metadata": null,
        "parts": [
          {
            "kind": "text",
            "metadata": null,
            "text": "MISSING_INFO: Please provide the date of the transaction."
          }
        ],
        "referenceTaskIds": null,
        "role": "agent",
        "taskId": null
      }
    ],
    "id": "9224947643702-7674c-417b-a0b0-f0741243c589",
    "kind": "task",
    "metadata": null,
    "status": {
      "message": null,
      "state": "completed",
      "timestamp": "2025-11-10T09:05:15.488894"
    }
  }
}
```

--------------------------------

### Send a request to the agent

Source: https://github.com/restatedev/ai-examples/blob/main/openai-agents/examples/langfuse/README.md

Send a request to the agent service running on Restate.

```bash
curl localhost:8080/InsuranceClaimAgent/run \
  --json '{"text": "Customer ID: cus_123 - Hospital bill for broken leg treatment at General Hospital for 3000 euro on 24/04/26"}'
```

--------------------------------

### Build MCP Server

Source: https://github.com/restatedev/ai-examples/blob/main/mcp/README.md

Commands to navigate to the restate-mcp directory, install dependencies, and build the server.

```shell
cd restate-mcp
npm install
npm run build

```

--------------------------------

### Environment Variables

Source: https://github.com/restatedev/ai-examples/blob/main/openai-agents/examples/arize_phoenix/README.md

Setting environment variables for API keys and collector endpoints.

```bash
echo 'OPENAI_API_KEY=sk-proj-...' > .env
echo 'PHOENIX_COLLECTOR_ENDPOINT=https://app.phoenix.arize.com/s/your-account-name' >> .env
echo 'PHOENIX_API_KEY=...' >> .env
echo 'PHOENIX_PROJECT_NAME=...' >> .env
```

--------------------------------

### Add API keys to .env file

Source: https://github.com/restatedev/ai-examples/blob/main/pydantic-ai/examples/logfire/README.md

Creates or appends API keys for Logfire and OpenAI to the .env file.

```bash
echo 'LOGFIRE_WRITE_TOKEN=your-write-token' >> .env
echo 'OPENAI_API_KEY=sk-proj-...' >> .env
```

--------------------------------

### Run the agent

Source: https://github.com/restatedev/ai-examples/blob/main/typescript-restate-only/template/README.md

Exports the OpenAI API key and starts the development server for the agent.

```bash
export OPENAI_API_KEY=sk-...
npm run dev
```

--------------------------------

### Register Services with Restate Server

Source: https://github.com/restatedev/ai-examples/blob/main/vercel-ai/nextjs-template/README.md

Registers the agent services with the Restate Server to enable durable execution.

```shell
npx @restatedev/restate deployments register -y --use-http1.1 http://localhost:3000/restate/v1
```

--------------------------------

### Send Request to Agent

Source: https://github.com/restatedev/ai-examples/blob/main/vercel-ai/nextjs-template/README.md

Sends a request to the agent via the Restate Server endpoint.

```shell
curl localhost:8080/agent/run --json '"What is the weather in Detroit?"'
```

--------------------------------

### Export OpenAI API Key

Source: https://github.com/restatedev/ai-examples/blob/main/openai-agents/tour-of-agents/README.md

Export your OpenAI API key before running the examples.

```bash
export OPENAI_API_KEY=your-key
```

--------------------------------

### Send Message (Happy Path)

Source: https://github.com/restatedev/ai-examples/blob/main/python-restate-only/examples/interrupt-regenerate/README.md

Curl command to send a message to the agent for the happy path scenario.

```bash
curl localhost:8080/CodingAgent/alice/message \
  --json '{"content":"Write me a small todo CLI in Python."}'
```

--------------------------------

### Interrupt with New Context

Source: https://github.com/restatedev/ai-examples/blob/main/python-restate-only/examples/interrupt-regenerate/README.md

Curl command to interrupt the ongoing task with new context.

```bash
curl localhost:8080/CodingAgent/bob/message \
  --json '{"content":"Actually, use FastAPI instead of Flask."}'
```

--------------------------------

### Generate a greeting tool

Source: https://github.com/restatedev/ai-examples/blob/main/mcp/README.md

An example tool that generates a greeting for a person, including creating a greeting card image and an audio file. It utilizes Restate's `ctx.run` for executing tasks with retry capabilities and `all` for parallel execution.

```typescript
tool(
  {
    description: "Greets a person with a song and dance",
    input: z.object({ name: z.string() }),
  },
  async (ctx, { name }) => {

    const urls = await ctx.run(
      "Obtain two Pre-signed URLs for a bucket",
      () => generatePresignedUrls()
    );

    const imageStep = ctx.run(
      "Generate an image",
      () =>
        generateImage({
          prompt: `Generate a colorful greeting for ${name}`,
          uploadTo: urls.imageUrl,
        }),
      {
        maxRetryAttempts: 3,
      }
    );

    const audioStep = ctx.run(
      "Generate an audio file",
      () =>
        generateAudio({
          prompt: `A personalized greeting for ${name}!`,
          uploadTo: urls.audioUrl,
        }),
      { maxRetryAttempts: 3 }
    );

    
    await all([imageStep, audioStep]);

    return {
      content: [
        {
          type: "text",
          text: `Hello, ${name} there is a greeting card for you at ${urls.imageUrl} and a song ${urls.audioUrl}}!`,
        },
      ],
    };
  }
);

```

--------------------------------

### Send First Message (Interruption Path)

Source: https://github.com/restatedev/ai-examples/blob/main/python-restate-only/examples/interrupt-regenerate/README.md

Curl command to send the first message in the interruption path (fire-and-forget).

```bash
curl localhost:8080/CodingAgent/bob/message/send \
  --json '{"content":"Build a Flask app with user auth."}'
```

--------------------------------

### Register the AI SDK agents at Restate

Source: https://github.com/restatedev/ai-examples/blob/main/vercel-ai/nextjs-example-app/README.md

Command to register the AI SDK agents with Restate using HTTP/1.1.

```shell
npx @restatedev/restate deployments register http://localhost:3000/restate/v1 --use-http1.1
```

--------------------------------

### Get Task Details

Source: https://github.com/restatedev/ai-examples/blob/main/a2a/README.md

This code snippet shows how to use curl to send a JSONRPC request to the ReimbursementAgentA2AServer to get task details.

```shell
curl localhost:8080/ReimbursementAgentA2AServer/process_request     --json '{
      "jsonrpc": "2.0",
      "id": 2,
      "method":"tasks/get",
      "params": {
        "id": "0d9g9ea6-dcc67-43ee-a389-8s9g0e6es5554",
        "history_length": 10,
        "metadata": {}
      }
    }' | jq .
```

--------------------------------

### Send a request to the agent

Source: https://github.com/restatedev/ai-examples/blob/main/python-restate-only/examples/langfuse/README.md

Sends a JSON request to the agent service running on Restate to process a reimbursement claim.

```bash
curl localhost:8080/ClaimReimbursement/process \
  --json '{"message": "Process my hospital bill of 2024-10-01 for 3000USD for a broken leg at General Hospital."}'
```

--------------------------------

### Send requests to the agent

Source: https://github.com/restatedev/ai-examples/blob/main/langchain-python/template/README.md

Send a request to the agent to get the weather in San Francisco.

```shell
curl localhost:8080/agent/run --json '{"message": "What is the weather in San Francisco?"}'
```

--------------------------------

### Get Agent History

Source: https://github.com/restatedev/ai-examples/blob/main/typescript-restate-only/examples/interrupt-regenerate/README.md

Retrieves the conversation history for an agent.

```bash
curl localhost:8080/CodingAgent/alice/getHistory
```

--------------------------------

### Register Services with Restate

Source: https://github.com/restatedev/ai-examples/blob/main/vercel-ai/template/README.md

Register the agent's services with the Restate Server to enable Restate Server to proxy invocations to the agent, adding durable execution.

```shell
npx @restatedev/restate -y deployments register localhost:9080
```

--------------------------------

### Coding Task Python Code

Source: https://github.com/restatedev/ai-examples/blob/main/python-restate-only/examples/interrupt-regenerate/README.md

This snippet shows the Python code for the CodingTask, which performs LLM steps (plan, draft, polish) and handles cancellation signals.

```python
import json
from typing import Any

from "restate.aio.service" import Service, ServiceContext
from "restate.exceptions" import TerminalError


class CodingTask(Service):
    async def run_task(self, ctx: ServiceContext, prompt: str) -> str:
        try:
            # Simulate LLM steps
            plan = await self.llm_plan(ctx, prompt)
            draft = await self.llm_draft(ctx, plan)
            polished_code = await self.llm_polish(ctx, draft)
            return polished_code
        except TerminalError:
            # Notify orchestrator and re-raise
            await ctx.send_event("task_cancelled", {})
            raise

    async def llm_plan(self, ctx: ServiceContext, prompt: str) -> str:
        # Simulate LLM call for planning
        await ctx.sleep(1)
        print(f"LLM Plan: {prompt}")
        return f"plan for {prompt}"

    async def llm_draft(self, ctx: ServiceContext, plan: str) -> str:
        # Simulate LLM call for drafting
        await ctx.sleep(1)
        print(f"LLM Draft: {plan}")
        return f"draft based on {plan}"

    async def llm_polish(self, ctx: ServiceContext, draft: str) -> str:
        # Simulate LLM call for polishing
        await ctx.sleep(1)
        print(f"LLM Polish: {draft}")
        return f"polished code from {draft}"

```

--------------------------------

### Send Request to Agent

Source: https://github.com/restatedev/ai-examples/blob/main/google-adk/template/README.md

Send a request to the agent to get the weather information for a city.

```shell
curl localhost:8080/agent/run --json '{
  "message": "What is the weather like in San Francisco?",
  "user_id": "user-123"
}'
```

--------------------------------

### Agent Python Code

Source: https://github.com/restatedev/ai-examples/blob/main/python-restate-only/examples/interrupt-regenerate/README.md

This snippet shows the Python code for the CodingAgent, which handles messages, manages conversation state, and interacts with the CodingTask.

```python
from typing import List, Tuple

from "restate.aio.virtual_object" import VirtualObject, HandledContext
from "restate.aio.service" import Service, ServiceContext
from "restate.aio.types" import Secret
from "restate.exceptions" import TerminalError

from coding_task import CodingTask


class CodingAgent(VirtualObject):
    def __init__(self) -> None:
        self.conversation_history: List[Tuple[str, str]] = []
        self.current_task_invocation_id: str | None = None

    async def message(self, ctx: HandledContext, content: str) -> None:
        if self.current_task_invocation_id:
            # Cancel the ongoing task if there is one
            await ctx.cancel_invocation(self.current_task_invocation_id)
            self.conversation_history.append(("user", "<interrupted>\n" + content))
        else:
            self.conversation_history.append(("user", content))

        task = CodingTask()
        invocation_id = await ctx.run_typed_invocation(task.run_task, content)
        self.current_task_invocation_id = invocation_id

        # We don't await the task here, it runs in the background
        # The invocation ID is stored so we can cancel it later

    async def get_history(self, ctx: HandledContext) -> List[Tuple[str, str]]:
        return self.conversation_history


class CodingAgentService(Service):
    async def message(self, ctx: ServiceContext, agent_id: str, content: str) -> None:
        agent = await ctx.get_virtual_object(CodingAgent, agent_id)
        await agent.message(ctx.handled(), content)

    async def message_send(self, ctx: ServiceContext, agent_id: str, content: str) -> None:
        # Fire-and-forget for the first message
        agent = await ctx.get_virtual_object(CodingAgent, agent_id)
        ctx.background_invocation(agent.message(ctx.handled(), content))

    async def get_history(self, ctx: ServiceContext, agent_id: str) -> List[Tuple[str, str]]:
        agent = await ctx.get_virtual_object(CodingAgent, agent_id)
        return await agent.get_history(ctx.handled())

```

--------------------------------

### Start a new reimbursement task

Source: https://github.com/restatedev/ai-examples/blob/main/a2a/README.md

This curl command initiates a new reimbursement task by sending a message to the ReimbursementAgentA2AServer.

```shell
curl localhost:8080/ReimbursementAgentA2AServer/process_request \
    --json '{ \
  "jsonrpc": "2.0", \
  "id": 1424443, \
  "method": "message/send", \
  "params": { \
    "message": { \
      "role": "user", \
      "parts": [ \
        { \
          "kind": "text", \
          "text": "Reimburse my hotel for my business trip of 5 nights for 1200USD of 05/04/2025" \
        } \
      ], \
      "messageId": "92249e73702-7674c-417b-a0b0-f0741243c449", \
      "taskId": "33349e733702-7674c-417b-a0b0-f0741243c333" \
    }, \
    "metadata": {} \
  } \
}' | jq .
```

--------------------------------

### Send a Request to the Agent

Source: https://github.com/restatedev/ai-examples/blob/main/vercel-ai/template/README.md

Send a request to the agent. The request targets Restate Server's endpoint (8080) because the server proxies requests to the service, making them durable.

```shell
curl localhost:8080/agent/run --json '{"prompt": "What is the weather in San Francisco?"}'
```

--------------------------------

### Start a task that blocks on human approval

Source: https://github.com/restatedev/ai-examples/blob/main/a2a/README.md

This command initiates a reimbursement request and requires human approval, demonstrating the initial step in a multi-turn conversation.

```shell
curl localhost:8080/ReimbursementAgentA2AServer/process_request \
    --json '{ \
  "jsonrpc": "2.0", \
  "id": 1424643, \
  "method": "message/send", \
  "params": { \
    "message": { \
      "role": "user", \
      "parts": [ \
        { \
          "kind": "text", \
          "text": "Reimburse my hotel for my business trip of 5 nights for 1200USD" \
        } \
      ], \
      "taskId": "0d9g9ea6-dcc67-43ee-a389-8s9g0e6es5554", \
      "contextId": "5e00b4a6-dcc67-43ee-a389-0e2a65958444", \
      "messageId": "9224947643702-7674c-417b-a0b0-f0741243c589" \
    }, \
    "metadata": {} \
  } \
}' | jq .
```

--------------------------------

### Get Interrupted Agent History

Source: https://github.com/restatedev/ai-examples/blob/main/typescript-restate-only/examples/interrupt-regenerate/README.md

Retrieves the conversation history after an interruption.

```bash
curl localhost:8080/CodingAgent/bob/getHistory
```

--------------------------------

### Register the Services

Source: https://github.com/restatedev/ai-examples/blob/main/typescript-restate-only/tour-of-agents/README.md

Register the services with Restate. Use --force if you already had another deployment registered at 9080.

```shell
restate -y deployments register localhost:9080 --force
```

--------------------------------

### Run an agent

Source: https://github.com/restatedev/ai-examples/blob/main/google-adk/tour-of-agents/README.md

Run a specific agent.

```bash
uv run app/chat_agent.py
```

--------------------------------

### Register the deployment

Source: https://github.com/restatedev/ai-examples/blob/main/google-adk/tour-of-agents/README.md

Register the Restate deployment.

```bash
curl localhost:9070/deployments --json '{"uri": "http://host.docker.internal:9080"}'
```

--------------------------------

### Run Restate Server

Source: https://github.com/restatedev/ai-examples/blob/main/typescript-restate-only/examples/braintrust/README.md

Run Restate server with the configuration file.

```shell
restate-server --config-file restate.toml
```

--------------------------------

### Register Services

Source: https://github.com/restatedev/ai-examples/blob/main/mcp/README.md

Command to register the services with the Restate server. Use `--force` if a deployment already exists.

```shell
restate -y deployments register localhost:9080

```

--------------------------------

### Cancel the reimbursement task

Source: https://github.com/restatedev/ai-examples/blob/main/a2a/README.md

This curl command cancels the previously started reimbursement task using its task ID.

```shell
curl localhost:8080/ReimbursementAgentA2AServer/process_request \
    --json '{ \
      "jsonrpc": "2.0", \
      "id": 3, \
      "method":"tasks/cancel", \
      "params": { \
        "id": "33349e733702-7674c-417b-a0b0-f0741243c333" \
      } \
    }' | jq .
```

--------------------------------

### Export Google API Key

Source: https://github.com/restatedev/ai-examples/blob/main/google-adk/template/README.md

Export your Google API key as an environment variable. Obtain an API key from Google AI Studio.

```shell
export GOOGLE_API_KEY=your_google_api_key_here
```

--------------------------------

### Claude Desktop Configuration

Source: https://github.com/restatedev/ai-examples/blob/main/mcp/README.md

JSON configuration for Claude desktop to specify the MCP server command and arguments.

```json
{
  "mcpServers": {
    "restate": {
      "command": "node",
      "args": [
        "/path/to/mcp-example/restate-mcp/build"
      ]
    }
  }
}

```

--------------------------------

### Export OpenAI API key

Source: https://github.com/restatedev/ai-examples/blob/main/langchain-python/template/README.md

Export your OpenAI or Anthropic API key as an environment variable.

```shell
export OPENAI_API_KEY=your_openai_api_key
```

--------------------------------

### Export API Key

Source: https://github.com/restatedev/ai-examples/blob/main/mcp/README.md

Command to export the OpenAI or Anthropic API key as an environment variable.

```shell
export OPENAI_API_KEY=your_openai_api_key

```

--------------------------------

### Export API Keys and Run Agent

Source: https://github.com/restatedev/ai-examples/blob/main/typescript-restate-only/examples/braintrust/README.md

Export necessary API keys and run the agent.

```bash
export BRAINTRUST_API_KEY=sk-...
export BRAINTRUST_PROJECT_ID=...
export OPENAI_API_KEY=sk-...
npm run dev
```

--------------------------------

### Register Services

Source: https://github.com/restatedev/ai-examples/blob/main/google-adk/template/README.md

Register the services with the Restate CLI.

```shell
restate -y deployments register localhost:9080
```

--------------------------------

### Run Reimbursement Agent

Source: https://github.com/restatedev/ai-examples/blob/main/a2a/README.md

Command to run the reimbursement agent application using UV.

```shell
uv run app/reimbursement
```

--------------------------------

### Run Weather Agent

Source: https://github.com/restatedev/ai-examples/blob/main/a2a/README.md

Command to run the weather agent application using UV.

```shell
uv run app/weather
```

--------------------------------

### Send Message to Agent

Source: https://github.com/restatedev/ai-examples/blob/main/python-restate-only/template/README.md

Send a message to the agent.

```shell
curl localhost:8080/agent/run --json '{"message":"What is the weather in San Francisco?"}'
```

--------------------------------

### Competitive Racing Agents

Source: https://github.com/restatedev/ai-examples/blob/main/README.md

Run parallel agents and use the fastest response, canceling others.

```python
python-restate-only/tour-of-agents/app/racing_agents.py
```

```typescript
typescript-restate-only/tour-of-agents/src/racing-agents.ts
```

--------------------------------

### Restate Configuration

Source: https://github.com/restatedev/ai-examples/blob/main/typescript-restate-only/examples/braintrust/README.md

Restate configuration file with Braintrust tracing endpoint and headers.

```toml
tracing-endpoint="otlp+https://api.braintrust.dev/otel/v1/traces"

[tracing-headers]
authorization="Bearer your-braintrust-api-key"
x-bt-parent="project_id:your-project-id"
```

--------------------------------

### Send First Message for Interruption

Source: https://github.com/restatedev/ai-examples/blob/main/typescript-restate-only/examples/interrupt-regenerate/README.md

Sends the initial message to the agent, fire-and-forget.

```bash
curl localhost:8080/CodingAgent/bob/message/send \
  --json '{"content":"Build a Fastify app with user auth."}'
```

--------------------------------

### Provide the date of the transaction

Source: https://github.com/restatedev/ai-examples/blob/main/a2a/README.md

This command continues the conversation by providing the missing transaction date, demonstrating how to respond to agent requests for more information.

```shell
curl localhost:8080/ReimbursementAgentA2AServer/process_request \
    --json '{ \
  "jsonrpc": "2.0", \
  "id": 1423, \
  "method": "message/send", \
  "params": { \
    "message": { \
      "role": "user", \
      "parts": [ \
        { \
          "kind": "text", \
          "text": "The date of the transaction is 05/04/2025" \
        } \
      ], \
      "taskId": "0d9g9ea6-dcc67-43ee-a389-8s9g0e6es5554", \
      "contextId": "5e00b4a6-dcc67-43ee-a389-0e2a65958444", \
      "messageId": "92249e73702-7674c-417b-a06b0-5sg6g881243c589" \
    }, \
    "metadata": {} \
  } \
}' | jq .
```

--------------------------------

### Register the agent

Source: https://github.com/restatedev/ai-examples/blob/main/typescript-restate-only/template/README.md

Registers the local agent with the Restate server.

```bash
restate deployments register http://localhost:9080 --force --yes
```

--------------------------------

### Agent output for initial request

Source: https://github.com/restatedev/ai-examples/blob/main/a2a/README.md

This JSON output shows the agent's response to the initial reimbursement request, indicating that it requires more information (the transaction date).

```json
{
  "id": 1424643,
  "jsonrpc": "2.0",
  "result": {
    "artifacts": null,
    "contextId": "5e00b4a6-dcc67-43ee-a389-0e2a65958444",
    "history": [
      {
        "contextId": "5e00b4a6-dcc67-43ee-a389-0e2a65958444",
        "extensions": null,
        "kind": "message",
        "messageId": "9224947643702-7674c-417b-a0b0-f0741243c589",
        "metadata": null,
        "parts": [
          {
            "kind": "text",
            "metadata": null,
            "text": "Reimburse my hotel for my business trip of 5 nights for 1200USD"
          }
        ],
        "referenceTaskIds": null,
        "role": "user",
        "taskId": "0d9g9ea6-dcc67-43ee-a389-8s9g0e6es5554"
      }
    ],
    "id": "9224947643702-7674c-417b-a0b0-f0741243c589",
    "kind": "task",
    "metadata": null,
    "status": {
      "message": {
        "contextId": null,
        "extensions": null,
        "kind": "message",
        "messageId": "89194646-702b-4af3-86d5-7c60355fbb77",
        "metadata": null,
        "parts": [
          {
            "kind": "text",
            "metadata": null,
            "text": "MISSING_INFO: Please provide the date of the transaction."
          }
        ],
        "referenceTaskIds": null,
        "role": "agent",
        "taskId": null
      },
      "state": "input-required",
      "timestamp": "2025-11-10T08:57:25.462064"
    }
  }
}
```

--------------------------------

### Happy Path Message

Source: https://github.com/restatedev/ai-examples/blob/main/typescript-restate-only/examples/interrupt-regenerate/README.md

Sends a message to the agent for a full plan/draft/polish cycle.

```bash
curl localhost:8080/CodingAgent/alice/message \
  --json '{"content":"Write me a small todo CLI in TypeScript."}'
```

--------------------------------

### Interrupt & Regenerate

Source: https://github.com/restatedev/ai-examples/blob/main/README.md

Interrupt a running agent with new context; cancel in-flight task and restart.

```python
python-restate-only/examples/interrupt-regenerate/agent.py
```

```typescript
typescript-restate-only/examples/interrupt-regenerate/src/agent.ts
```

--------------------------------

### Agent logs indicating reimbursement approval

Source: https://github.com/restatedev/ai-examples/blob/main/a2a/README.md

These log snippets show the agent's internal process, including calling an LLM and executing the 'reimburse' tool, culminating in a curl command to approve the request.

```text
... first part of logs ...
[2025-05-16 13:42:50,410] [310993] [INFO] - Agent session lwp13w5e3sdf258t3wesf13234 -   Starting iteration of agent loop with agent: ReimbursementAgent and tools/handoffs: ['create_request_form', 'reimburse', 'return_form']
[2025-05-16 13:42:50,410] [310993] [INFO] - Agent session lwp13w5e3sdf258t3wesf13234 -  Calling LLM
[2025-05-16 13:42:52,293] [310993] [INFO] - HTTP Request: POST https://api.openai.com/v1/responses "HTTP/1.1 200 OK"
[2025-05-16 13:42:52,303] [310993] [INFO] - Agent session lwp13w5e3sdf258t3wesf13234 -  Executing tool reimburse
==================================================
 Requesting approval for request_id_1633297
 Resolve via:
curl localhost:8080/restate/awakeables/sign_1oqmHpDF_RJQBltjnf48zszmfmRr4w9izAAAAEQ/resolve --json '{"approved": true}'
```