Skip to content

Executor vs. Agent Client

Executor and Agent Client can both run on a user’s computer or inside a private network. They solve different problems, plan work in different places, and use different identities.

Dimension Agent Client Executor
Main purpose Let a local agent discover and invoke limited capabilities as an authorized business subject Let a local or private runtime claim a Job that BailingHub has already created
Who plans The local agent The Hub has already selected a target and dispatched the task
Identity A revocable Agent Session created by browser authorization and bound to a business subject and allowed routes A target-scoped Executor Token that identifies a worker node, not an end-user business identity
Interaction Ongoing local conversation; DSH fixes an explicitly selected scope per conversation The worker claims, heartbeats, executes, and reports by Job ID
Capability source A restricted workspace projection of tools and rules The execution target configuration and task envelope
Typical use A desktop agent operates orders, staff, or tickets Code review, local files, private scripts, or a specialized runtime

The user wants to state a goal in a local agent and let the host agent choose multiple business tools. The user confirms identity on a business-side authorization page. BailingHub still enforces capability projection, approval, idempotency, invocation, and audit.

DSH requires scope selection before the first message; an empty or unset scope means ordinary chat. Authorizations with different clientAppId + workspace bindings in one Hub and audit domain can be selected per action. Losing any selected authorization blocks the entire scope.

Accounts and conversation records explains synchronized visible text, per-turn execution links, missing history, and retrying uploads after reconnecting. Upload retries do not repeat business execution or turn execution summaries into original conversation text.

BailingHub has already created a task that must run in a local or private environment the Hub cannot reach directly. An Executor is a task worker. It does not inherit a browser login session, and its token must not be treated as authority to represent any business user.

  • Tokens: an Agent access token and an Executor Token have different meanings.
  • Configuration: an Agent Client connection/workspace is not an execution target.
  • Audit subjects: one records an authorized business identity; the other records the worker that claimed a task.
  • Authorization: an Executor Token cannot replace the business subject or final authorization required by a tool call.