A hands-on tour of VeADK, from the smallest possible agent to a multi-agent workflow. Each folder is self-contained: a single main.py, an .env.example, and a bilingual README (English + 中文).

中文版见 README.zh.md

There are also frontend-focused demos that run with veadk frontend --agents-dir examples:

• a2ui_agent/ demonstrates agent-driven UI.
• multimodal_agent/ analyzes images, TXT/Markdown, PDFs, and videos uploaded from the chat composer.

For a deployable full app (web UI + agent API in one container, shipped to Volcengine AgentKit via veadk agentkit), see basic-app/.

The examples are grouped by concept: 01–02 basics, 03 & 09 memory, 04–05 tools & knowledge, 06 & 10 multi-agent, 07–08 model behavior, 11 observability, and 13 OpenViking-backed knowledge and memory.

Four examples use Agent(runtime="codex"), which hands the inner loop to a sandboxed coding agent that can write a file, run it, read the traceback and fix it. Two of them show what the runtime is for; two show how to wire it up.

Reach for this runtime when the steps cannot be enumerated as tools ahead of time — ad-hoc analysis, log triage, data wrangling. Keep runtime="adk" for a fixed tool call and a formatted answer: it is faster, cheaper, and does not reject sub_agents / output_schema / planner / code_executor. See when to use the codex runtime.

1. Install VeADK (example 05 with the local backend needs the extensions extra):

   pip install veadk-python
   # for example 05's local RAG backend:
   pip install "veadk-python[extensions]"

2. In each example folder, copy the env template and add your keys:

   cd 01_quickstart
   cp .env.example .env

   VeADK auto-loads .env from the current working directory. You can also use a config.yaml instead — see the configuration docs.

3. Run:

   python main.py

• Agent — model + instruction + tools + memory/knowledge. Created once, reads model config from the environment.
• Runner — drives a conversation; await runner.run(messages=..., session_id=...) returns the final text answer.
• Tools — any Python function with type hints and a docstring; built-ins live under veadk.tools.builtin_tools.
• Memory — short-term (conversation context, keyed by session_id) and long-term (across sessions).
• KnowledgeBase — RAG over your documents; auto-adds a retrieval tool.
• Workflow agents — SequentialAgent, ParallelAgent, LoopAgent to orchestrate multiple agents.

• Docs: https://volcengine.github.io/veadk-python/
• Tutorial notebook: veadk_tutorial.ipynb