KsADK

Why KsADK

KsADK is not another agent framework. It is a unified runtime layer for agents built with existing frameworks.

Most frameworks answer "How do I build an agent?". KsADK focuses on "How do I run, debug, expose, deploy, and observe agents consistently?"

One-line positioning

Kingsoft Cloud Agent Development Kit: a cloud-native framework to build, deploy, debug, and observe enterprise AI agents.

Keep building business logic with Google ADK, LangGraph, LangChain, or DeepAgents. Use KsADK for one CLI, local Web UI, OpenAI-Compatible API, Skill Runtime, Workspace, Sandbox, OpenClaw / Hermes runtime launch paths, and OpenTelemetry observability.

Relation To Other Frameworks

Google ADK focuses on agent modeling, tools, multi-agent collaboration, Session/Memory, local runs, and Web debugging. ADK solves agent development; KsADK solves agent runtime.

KsADK does not replace ADK. It lets ADK agents share the same local runtime, browser debugging UI, OpenAI-Compatible API, and deployment entrypoint as LangGraph, LangChain, and DeepAgents projects. Useful when: your team has both ADK and LangGraph projects, you want one Web UI to debug agents from different frameworks, you need to expose local agents through OpenAI-Compatible APIs, or you need Kingsoft Cloud AgentEngine/Skill/Workspace/Sandbox/deployment integration.

LangGraph is strong at graph-state orchestration, checkpointing, streaming, human-in-the-loop workflows, and the LangChain ecosystem.

KsADK does not rewrite LangGraph execution. It adds a runtime platform layer around LangGraph projects: agentengine run for local terminal interaction, agentengine web for browser debugging, /v1/responses and /v1/chat/completions local protocols, Skill Runtime/Workspace/Sandbox toolsets, and AgentEngine/Hermes/OpenClaw/Serverless deployment entrypoints.

OpenAI Agents SDK is native to the OpenAI Responses API and provides agent orchestration, tool calling, handoffs, guardrails, and tracing.

KsADK is multi-framework and multi-runtime. It does not require your business agent to be written with the OpenAI Agents SDK. It wraps ADK, LangGraph, LangChain, and DeepAgents projects into one local runtime, debugging, OpenAI-Compatible API, and deployment workflow.

What KsADK Solves

Start fast

`agentengine init` / `agentengine config` / `agentengine run` to launch new agent projects.

Browser debugging

`agentengine web` and the ksadk-web UI to debug streaming, attachments, and tool calls.

Unified protocol

`/v1/responses`, `/v1/chat/completions` consistent across frameworks.

Tools and isolation

Skill Runtime, Workspace tools, Sandbox tools.

Deploy remotely

`agentengine build`, `agentengine launch`, Hermes/OpenClaw.

Unified observability

OpenTelemetry / OTLP tracing.

When You May Not Need KsADK

If you only have a single-file script that directly calls a model API and you do not need browser debugging, isolated tools, unified protocols, deployment, or observability, the framework itself may be enough. KsADK becomes useful when you need one way to manage multiple agent projects.

On this page