Context search and Agent tools: from discovery to use

A work scope may contain many conversations, files, and deliverables. The platform separates finding materials, choosing references, and reading their content during an AgentRun so each request uses the context it needs.

A shared work scope is not one shared conversation

Workspaces and work items organize related work, but each conversation keeps its own message history. Other files and conversations in the same scope are available to discover; they are not automatically inserted into every Agent request. This lets a scope retain a long history without turning every new question into a large, undifferentiated prompt.

Search locates resources before content is read

Context search helps people find conversations, files, deliverables, and work items by scope and result type. Results carry identifiers, ownership, and limited preview information so a person can open the right item. The Agent has a separate scoped resource-search tool for resource metadata. A search result is a lead to inspect, not a claim that the complete resource has already entered the model context.

A reference makes the intended material explicit

After opening a relevant file or deliverable, a person can add it as a reference in the AI panel and review that selection before sending. The reference identifies the material and its owning scope; it does not move the original file or grant access to a different workspace. This separates finding something useful from asking the Agent to use it for a particular request.

Tools resolve content within the execution boundary

When an AgentRun needs an uploaded file or artifact, the Agent can resolve its identifier into a readable item through a scoped tool. Materials owned by another scope in the same workspace are made available within the current run's working area when permitted. The Agent then reads the material it needs instead of receiving every matching resource as part of the initial prompt.

Discovery and access remain separate decisions

The human search interface, the Agent's metadata search, and content-resolution tools serve different purposes. Each operates within its own scope and permission rules. Keeping these steps separate makes it easier to inspect where a result came from, choose what to reference, and keep Agent work within the authorized workspace.

Continue reading