Visual AI workflows and agent orchestration
ChengOS expresses an agent as a directed acyclic graph: each node is one inspectable action, each edge is data. That makes a process testable, reusable and shippable instead of buried in a prompt.
Nodes and sub-workflows
A workflow is built from nodes whose inputs and outputs are described by JSON Schema on the backend; the editor turns that schema into form controls. Any section of a workflow can be packaged as a sub-workflow and reused elsewhere as a single node.
ReAct agent nodes
Alongside fixed orchestration, a workflow can contain ReAct-style agent nodes where the model decides which tools to call and how many rounds to take. Mixing the two lets you trade determinism against flexibility per step.
From debugging to release
Workflows can be stepped through with intermediate results visible, then released once they behave. Execution is exposed over REST and WebSocket so external systems can trigger runs and subscribe to progress.
Credentials stay out of the model
Calls to external services use credential handles, so secrets never appear in the model context. It is a platform-level guarantee rather than something each workflow has to re-implement.
Questions people ask
- Can I try it before installing anything?
- Yes — the home page has a read-only gallery of product screenshots. It does not offer an interactive editor or workflow execution. To use the product, install the Windows client or deploy ChengOS yourself.
- Can external systems trigger a workflow?
- Yes, execution is exposed over REST and WebSocket. See the documentation site for the exact endpoints and parameters.
Evidence for the claims on this page
Last verified:
- 支持 MCP 的 Stdio、SSE 与 HTTP 三种传输方式。
- Scope: ChengOS 平台侧 MCP 客户端。
- Source: ChengOS MCP 传输实现 ·
- 模型只能通过凭据句柄调用工具，API Key 明文不进入模型上下文。
- Scope: 凭据保管库与工具调用路径；不涵盖用户自行粘贴到对话中的内容。
- Source: ChengOS Credential Vault 与工具调用实现 ·