Skip to main content

Test with the playground

The hosted playground gives you a chat UI for your agent without writing any frontend code. You reach it through one command: agentflow play.

How it works​

agentflow play does two things at once:

  1. Starts the same local API server as agentflow api.
  2. Opens the hosted playground in your browser with your local backend URL pre-configured.

Your graph runs locally. The hosted playground is just a UI that calls your local API.

Start the playground​

From the folder that contains agentflow.json:

agentflow play --host 127.0.0.1 --port 8000

Expected output:

INFO: AgentFlow API starting on http://127.0.0.1:8000
INFO: Opening playground at https://playground.agentflow.dev?backendUrl=http://127.0.0.1:8000

The browser should open automatically. If it does not, copy the URL from the terminal output and open it manually.

What you can test​

In the playground chat UI:

  1. Send a message — type a question and press send. The playground calls POST /v1/graph/invoke on your local API.
  2. See tool calls — if your agent uses tools, the playground shows which tools were called and what they returned.
  3. Inspect raw messages — use the debug panel to see the full message array including tool role messages.
  4. Test multiple threads — create a new conversation to start a fresh thread_id.

Example messages to try with the agent from the previous pages:

What is the weather in Paris?
What about London?
What did I ask about first?

The last question tests memory — the agent should remember your earlier question because the checkpointer saved the conversation.

Thread IDs in the playground​

Each conversation the playground starts gets a generated thread_id. The playground passes it with every request so the checkpointer can restore context between messages.

Stop the playground​

Press Ctrl+C in the terminal to stop the API server. The hosted playground will lose its backend and show a connection error until the server is restarted.

What you learned​

  • agentflow play starts the API server and opens the hosted playground in one command.
  • The playground calls your local API — your graph never leaves your machine.
  • You can test multi-turn memory, tool calls, and raw message structure from the UI.

Next step​

Call the agent programmatically from a TypeScript application — Call from TypeScript.