Build a durable chat

This tutorial runs a local Agent Runtime API and connects the Durable Chat example through the public Go SDK. The API configuration uses in-memory storage, so it is for exploration rather than durable deployment.

Start the API

From the repository root, set two local credentials and run the API:

export AGENT_RUNTIME_ADMIN_TOKEN='replace-with-at-least-16-bytes'
export AGENT_RUNTIME_DEVELOPER_TOKEN='replace-with-at-least-16-bytes'
go run ./cmd/agent-runtime-api --config "$PWD/deploy/runtimeapi/api.example.json"

The API listens at http://127.0.0.1:8088. GET /healthz and GET /readyz confirm that the process is serving.

Open Durable Chat

In a second terminal, run the loopback web UI:

go run ./examples/durable-chat/cmd/durable-chat \
  --mode=web \
  --runtime-url=http://127.0.0.1:8088

Open the printed http://127.0.0.1:8090 address. Create a Session, send an input, and watch its ordered events. The browser never receives the runtime bearer token: the local UI owns the SDK client and accepts browser requests only from its own loopback origin.

Try the terminal instead

Stop the web UI or use another terminal and run:

go run ./examples/durable-chat/cmd/durable-chat \
  --mode=terminal \
  --runtime-url=http://127.0.0.1:8088

Use new <agent-revision> to start a Session, then send, resume, events, and cancel to explore the client workflow.

What this demonstrates

The example uses only the public HTTP/SDK surface. It creates or resumes a Session, submits idempotent input, reconnects to cursor-addressed events, and issues explicit cancellation. Model-provider execution and sandboxed Workspace execution are intentionally separate operator-owned capabilities.