Can Kosmo complete a long sequence without supervision?
It can chain dependent actions within the active mode, available tools, iteration limits, and approval boundaries. The exact current boundary is published in the live profile above.
Agent architecture
Kosmo works inside a Kosmico project. Its useful unit is not an isolated message, but a task carried out against project files, research context, tools, permissions, and an explicit operating mode.
The official profile below is the source for changeable behavior. It distinguishes what the agent can do, what needs permission, what differs by surface, and what is sent outside the device.
A long research request can involve dependent actions, but access to tools, write operations, approvals, iteration limits, and the selected mode constrain the run. A capability shown in one surface should not be assumed to exist identically in every surface.
The profile therefore publishes modes and surface differences instead of using autonomy as a single marketing label.
Project context can include files, papers, notes, research-graph information, project memory, and recent work. Retrieval selects relevant material for the current task; the entire project is not blindly pasted into every model call.
Switching models changes the inference engine, not the project that owns the context. An explicitly selected model stays selected; the Recommended policy follows the current managed default.
A project file stored locally, a project enabled for synchronization, a cloud-model request, and a local-model request have different data flows. The profile states them separately so local-first is not misread as a promise that every operation is offline.
Official live source
The live product profile is temporarily unavailable. This page is withholding changeable product facts until the official source responds.
It can chain dependent actions within the active mode, available tools, iteration limits, and approval boundaries. The exact current boundary is published in the live profile above.
No. Relevant context is selected for the task. When cloud inference is used, the request and selected context are sent according to the cloud-model data flow.
The current BYOK status is a live product fact in the model policy above. It should not be inferred from the names of available models.