Open mode

Governed context is the right default for questions about your data. It is not the right tool for everything.

Sometimes you want an assistant that is simply capable: a general research question, a call into an internal service, a workflow that has nothing to do with the company numbers. Forcing that through a context scope adds ceremony without adding value.

Open mode is the other mode. It is deliberately unguarded by scope.

Custom agents

An agent bundles a model, its instructions, and a set of tools. You pick the model, write the instructions, choose the tools, and get an assistant configured for one workflow. Agents are first-class objects — you can create one per workflow and share it with other people.

See agents.

Any MCP server

The Model Context Protocol ecosystem is the tool belt. Databases, SaaS APIs, internal services — if it speaks MCP, an agent can call it. Attaching a new tool is a configuration step, not an integration project.

See MCP servers.

Caution

Open mode is open by design, which means it inherits whatever permission the attached tool has. An agent pointed at a writable endpoint can write. Keep open mode pointed at things that are safe to touch, and keep governed scopes for anything connected to production data.

Governed versus open

	Governed chat	Open mode
Context	One scope’s entities, metrics, and rules	Whatever tools the agent has
Bounded by	The scope’s context and access	Nothing — that is the point
Best for	Questions about your data	General assistance and custom workflows
SQL	Validated read-only before execution	Whatever the attached tool permits

Why both exist

Governance has a cost: scope setup, trust assignment, certification. That cost is worth paying for a number someone will act on, and it is pure overhead for everything else. Running both modes means the guarded path stays genuinely guarded — instead of the whole product being loose enough to be useful for side-tasks, the loose part is a separate, explicit mode.