A11 Guides

Run A11 examples for streams, tools and remote workers.

Start with one local action, then add model calls, tools, remote workers and Studio as the application needs them. Every guide ends in code you can run.

Start here

Stream a model reply and thoughts separately

Easy

Call a hosted model and return its visible reply and thoughts on separate text outputs.

ResultA prompt enters one action and leaves as separate reply and thought streams.promptactionreplythoughtsA Python action with separate reply and thought streams.

Design the application interface

Decide where streams, completion and application policy belong.

Add streaming and remote calls to an existing AI application

Intermediate

Add named streams and remote dispatch to one existing operation while application policy stays in familiar code.

Application policy calls one A11 action that can reach a model API or a remote worker.app policyactionmodel APIGPU workerA concrete integration point for adding A11 to an application.

Handle partial results, completion and cancellation in one action

Intermediate

Run one action whose caller receives partial values, then handles completion and cancellation separately.

A producer exposes early values and a separate completion status.producerstreamvaluesstatusA streamed action with explicit completion and cancellation.

Let an AI model update state inside a browser page

Advanced

Serve browser-owned state as typed actions, let a model call them through the existing session, and validate every change in the page.

A model calls typed actions through a session to inspect and change browser-owned state.page stateactionssessionmodelA live page where model tool calls inspect, recolour and move rendered objects.

Build model experiences

Follow audio, research and image generation through runnable code.

Transcribe a WAV file and read recognition events separately

Intermediate

Run local speech recognition, connect voice input to a streaming model, and follow the audio and text through the code.

Audio flows through recognition into editable text and a model call.ASRtextmodelA local WAV transcription with separate transcript and event streams.

Receive progress and a PNG from one image request

Intermediate

Run a real image model, watch progress arrive, and save the resulting PNG through a separate typed output.

One image request produces progress updates and a final PNG.promptmodelprogressPNGA generated PNG plus structured progress from one remote call.

Run parallel model research and stream the report to Python

Advanced

Run concurrent model investigations, read the plan and report through separate streams, and inspect the composition behind them.

One question starts investigations into API shape, failures and cancellation, then joins their findings in synthesis.questionAPI shapefailurescancellationsynthesisA real hosted investigation with a streamed plan and report.

Connect and inspect the system

Move work to another process and trace tool decisions in Studio.

Run an A11 action in a second process over WebSocket

Easy

Run one local worker and one caller over WebSocket while streamed results arrive in the caller.

A local caller streams results from an action on a remote worker.callerconnectionworkerOne worker and one caller communicating over a local WebSocket.

Trace a nested agent tool call through every outcome

Advanced

Run a nested tool workflow and inspect its arguments, streamed progress, result and final status.

A parent action exposes its nested tool call, result, logs and status.requestparenttoollogsstatusA reproducible nested tool call with success, failure and cancellation.