Python Streaming Cookbook

This directory collects Python examples that mirror the TypeScript workspace in ../typescript/. Most packages are backends — they run langgraph dev on port 2024 with the same graph wire-shape as their TypeScript sibling, so you can keep the existing React/Angular/Svelte/Vue frontends unchanged.

Start with streaming when learning protocol events and projections, then move to a UI backend when you want to exercise framework SDKs against Python.

Workspace Setup

Create the shared environment file from the repository root:

cp .env.example .env

Fill in the provider keys listed in .env. Every Python package loads this file (via langgraph.json or explicit load_dotenv in standalone servers).

Each example is an independent uv project. Install and run from the package directory:

cd python/<package>
uv sync

There is no shared Python workspace file — treat each folder like its own mini-repo. TypeScript frontends still use pnpm from ../typescript/ when an example pairs a Python backend with an existing UI.

Example Set

Package	Focus	TypeScript sibling	Start here
streaming	Terminal scripts for in-process and remote streaming.	typescript/streaming	Learn astream_events(..., version="v3"), projections, subgraphs, subagents, custom transformers, interrupts, and A2A streams.
multimodal	Six-way fan-out story graph (text, images, audio).	typescript/multimodal	Python backend only — run pnpm dev:multimodal for the React UI.
a2ui	ReAct agent with custom:a2ui projection.	typescript/a2ui	Python backend only — run pnpm dev:a2ui for the React UI.
ui-react	Reconnect / replay against langgraph dev.	typescript/ui-react	Python backend only — run pnpm dev:react for the React UI.
react-custom-backend	Full-stack custom Agent Protocol server + React UI.	typescript/react-custom-backend	LocalThreadSession, per-thread SSE, HttpAgentServerAdapter, tool streaming. No TypeScript backend required.
ui-angular	Minimal chat agent.	typescript/ui-angular	Python backend — run the Angular app from typescript/ui-angular.
ui-svelte	Minimal chat agent.	typescript/ui-svelte	Python backend — run the Svelte app from typescript/ui-svelte.
ui-vue	Minimal chat agent.	typescript/ui-vue	Python backend — run the Vue app from typescript/ui-vue.

Common Commands

Terminal streaming (python/streaming)

cd python/streaming
uv sync

# In-process
uv run python -m basic.in_process
uv run python -m messages.in_process
uv run python -m subagents.in_process

# Remote (spawns local langgraph dev, then connects via SDK)
uv run python -m basic.remote
uv run python -m custom_transformer.remote

See streaming/README.md for the full script map.

LangGraph dev backends (port 2024)

Most UI backends share the same run pattern:

cd python/ui-react   # or multimodal, a2ui, ui-angular, ui-svelte, ui-vue
uv sync
uv run langgraph dev --port 2024

Then start the matching frontend from typescript/ (see each package README for the exact pnpm command).

Full-stack custom backend (port 9123)

cd python/react-custom-backend
uv sync
uv run python src/main.py

# other terminal
cd python/react-custom-backend/frontend
npm install
npm run dev

SDK and Protocol Docs

Python examples use LangGraph's Python v3 event stream and, for remote scripts, langgraph-sdk. UI apps use the same @langchain/react (and other framework) packages as the TypeScript frontends.

• LangGraph Python docs
• Agent streaming protocol
• Python protocol bindings
• React SDK docs (frontends call the JS SDK against Python servers)

How To Choose An Example

• Choose streaming/basic when you need raw protocol events and namespaces.
• Choose streaming/messages when building token, reasoning, or tool-call UI.
• Choose streaming/custom-transformer or react-custom-backend when you need custom projections or your own HTTP/SSE backend.
• Choose a2ui when rendering declarative A2UI surfaces from a Python agent.
• Choose streaming/subgraphs or multimodal when scoped namespaces matter to the UI.
• Choose streaming/subagents when Deep Agents task delegation is the user-facing concept.
• Choose ui-react when testing browser refresh recovery.
• Choose react-custom-backend when implementing the Agent Protocol without langgraph dev.
• Choose a ui-* backend when validating framework bindings against a Python graph server.

Environment Notes

• python/streaming uses the Anthropic key from the root .env.
• Most UI backends and react-custom-backend use the OpenAI key.
• Remote streaming scripts need langgraph-sdk with client.threads.stream() support (see streaming/README.md).