Runnable Blueprints & Scaffolds

Root code-examples/

Production Code Scaffolds

11 tested, standalone engineering scaffolds and architectural blueprints demonstrating core AI engineering patterns — from multi-server MCP routing and LangGraph supervisors to Deep Agents composite backends and Docker GPU serving.

Total Blueprints11

Active Runnable7

Reference Designs4

Test Coverage100%

All Blueprints (11)Active Runnable (7)Architectural Designs (4)

Runtime:All Python Node.js Docker

Active Runnable

Agent Runtimes

Python 3.10+ (uv/pip)

Deep Agent Minimal Harness Blueprint

Deep Agents SDK with DeltaChannel state and CompositeBackend

A standalone, runnable Python reference implementation demonstrating the core architecture of Deep Agents: compiling an agent graph with DeepAgentState, DeltaChannel bounding checkpoint growth to O(N), CompositeBackend virtual storage multiplexing (/workspace to disk, /scratch to ephemeral memory), and custom audit middleware.

Architecture Layers:

Agent RuntimeVirtual BackendState Layer

Prerequisites:

•Python 3.10+
•uv or pip
•Anthropic API Key (ANTHROPIC_API_KEY)

Terminal Run Command:click to select all

uv pip install -e . && python src/main.py

Reference Repo System Teardown

code-examples/deepagent-minimal-harness

Active Runnable

Protocol & Transport

Python 3.10+

Multi-Server MCP Hub Scaffold

Concurrent stdio and SSE transport routing across heterogeneous tool providers

A production-grade multi-server Model Context Protocol (MCP) client scaffold that concurrently orchestrates tools, resources, and prompt templates across multiple independent MCP servers over stdio and SSE transports.

Architecture Layers:

Transport LayerProtocol IngressMulti-Server Router

Prerequisites:

•Python 3.10+
•pip or uv
•Model Context Protocol SDK

Terminal Run Command:click to select all

pip install -e . && python src/main.py

Reference Repo System Teardown

code-examples/mcp-multi-server-hub

Active Runnable

Multi-Agent Systems

Python 3.10+ / Docker

DeerFlow 2.0 SuperAgent Reference Scaffold

Multi-agent supervisor, dynamic SKILL.md parsing, and Docker container sandbox harness

Clean Python reference scaffold demonstrating the core architectural patterns of ByteDance DeerFlow 2.0: LangGraph Pregel lead supervisor, declarative SKILL.md tool loader, Docker container isolation, and durable SQLite state checkpointer.

Architecture Layers:

SupervisorWorker ClusterSandbox IsolationState Checkpointer

Prerequisites:

•Python 3.10+
•Docker Engine running
•pip install -r requirements.txt

Terminal Run Command:click to select all

python -m pip install -r requirements.txt && python main.py

Reference Repo System Teardown

code-examples/deerflow-superagent-scaffold

Active Runnable

Autonomous Coding Agents

Python 3.10+ / Docker

Open-SWE Minimal Runner Scaffold

Autonomous SWE agent execution loop with Docker sandbox and file patch isolation

Autonomous software engineering agent runtime inspired by SWE-bench and OpenHands. Implements the core observe-think-act loop inside isolated containerized development environments with unified diff patching.

Architecture Layers:

Agent LoopExecution SandboxDiff Patch Engine

Prerequisites:

•Python 3.10+
•Docker Engine running
•Git installed

Terminal Run Command:click to select all

python src/main.py

Reference Repo System Teardown

code-examples/open-swe-minimal-runner

Active Runnable

Web Ingress & Streaming

Node.js 18+ / Next.js 15

Next.js 15 Streaming Chat API

High-throughput token streaming with Vercel AI SDK, Server-Sent Events, and backpressure

Production Next.js 15 App Router streaming chat endpoint utilizing the Vercel AI SDK. Features robust backpressure handling, client-side abort controllers, tool calling chunk parsing, and real-time token rendering.

Architecture Layers:

Edge IngressStreaming TransportToken Stream Parser

Prerequisites:

•Node.js 18+
•npm / pnpm / yarn
•OpenAI or Anthropic API Key

Terminal Run Command:click to select all

npm install && npm run dev

Reference Repo

code-examples/streaming-api

Active Runnable

Inference & Model Serving

Docker / NVIDIA GPU

Dockerizing Ollama for GPU Inference

Containerized local LLM serving with NVIDIA Container Toolkit passthrough and health checks

Turnkey Docker Compose orchestration for self-hosting open-weight models (Llama 3.2, DeepSeek-R1, Qwen 2.5) with full NVIDIA GPU passthrough, persistent volume storage, and automated health checks.

Architecture Layers:

Container OrchestrationGPU Hardware PassthroughInference Runtime

Prerequisites:

•Docker Engine 24+
•Docker Compose v2
•NVIDIA Container Toolkit (nvidia-smi)

Terminal Run Command:click to select all

docker compose up -d && docker compose exec ollama ollama run llama3.2

Reference Repo System Teardown

code-examples/docker-ollama

Active Runnable

Multi-Agent Systems

Python 3.10+

LangGraph Supervisor Pattern

Multi-agent coordination with specialized researcher and coder subagents plus SQLite checkpointing

Reference implementation of the supervisor multi-agent architecture in LangGraph. A top-level supervisor delegates incoming user requests between a research agent and a coding agent, with SQLite checkpoint persistence.

Architecture Layers:

Supervisor NodeSubagent WorkersSQLite State Checkpointer

Prerequisites:

•Python 3.10+
•pip / uv
•OpenAI or Gemini API Key

Terminal Run Command:click to select all

pip install -r requirements.txt && python main.py

Reference Repo System Teardown

code-examples/langgraph-supervisor