Examples by task

Each guide starts from an application outcome and introduces the required A11 interfaces. Choose a streaming API, model service, agent, browser, or operations example below.

Build streaming APIs and pipelines

• Return values as they are produced. Write, read, finalize, and fail an asynchronous stream.
• Serve an action over WebSocket. Connect a client and server, send a call, and shut both sides down cleanly.
• Host the same operation locally or remotely. Move inference or data processing to another peer without changing its schema or port I/O.
• Compose available actions at runtime. Check and run a concurrent composition without deploying another handler.
• Stream progress and a finished image separately. Keep lightweight status records independent from a binary model result.
• Represent an HTTP exchange as streams. Process status, headers, body, trailers, and pushed responses as they arrive.

Add models and tools

• Stream a model response. Send a portable conversation to Claude, Gemini, or Ollama through one action interface.
• Let a model call an application action. Turn an action schema into a tool and run requested calls through its registry.
• Use tools published by an MCP server. Convert MCP declarations into actions that models, flows, and remote peers can call.
• Build a parallel research agent in Python. Plan, investigate several briefs concurrently, and stream the final report.
• Let a model compose its tools. Keep intermediate tool results out of model context by running their composition as one action.
• Build a harness or evaluation path. Assemble only the session, storage, policy, and result streams an application needs.

Build browser experiences

• Call an A11 service from a browser. Use the same action contract in a Python backend and a TypeScript page.
• Continue a chat after a reload. Store structured interactions and reopen a conversation by ID.
• Let a model use tools in the page. Serve actions over the page's existing connection so the model can operate on browser state.
• Run a model in the browser. Stream WebGPU inference through the same ports as a hosted model.

Distribute and operate

• Export runtime telemetry with OpenTelemetry. Configure OTLP or Langfuse and inspect action, session, and stream spans.
• Exchange durable streams through Redis. Let two programs communicate without a direct connection.
• Choose and verify the native allocator. Enable the packaged allocator for native executables or embedding processes.
• Integrate Flow diagnostics into an editor or CI. Read stable diagnostic, formatting, completion, and navigation responses.

Runnable companion programs are available in the repository's examples/ directory.