Products/AI agent controls
AI agent controls

Know what AI did.
Decide what it may do.

See decisions and handoffs, set the rules, and review what needs a person. Try an agent draft on the test bench before a separate deployment.

Silver knobs and sockets across a black modular control panel.

A clear role.
A visible decision.

Keep AI oversight close to the work: agent performance in Quality, held decisions in Approvals, and rules, skills, and drafts in Settings.

BearScope / AI agent controlsIllustrative records
Quality · Approvals · Settings

See the decision.
Set the boundary.

Shadow mode

Letra decisions

Flag a source that needs checking

Decision recorded in shadow. No action executed.

Read the decision

Inspect the available evidence, proposed response, and configured mode. A recorded decision and an executed action are different events.

Try a draft first.The test bench evaluates prompts with tools off. Saving a draft does not deploy an agent or enable customer messages, refunds, or fulfillment actions.

Current app workflow with fictional records. Features depend on connected sources and workspace access.

Light and shadow fall across a blue door and its chrome handle.

See the work it leaves behind.

Review AI handoffs and conversation evidence in Quality. Use the decision ledger to see shadow assessments, acted decisions, and requests passed to a person.

Set the boundary.

Configure decision thresholds, guardrails, and approved skills. Read access and permission to change a record are separate; only supported, enabled actions may run.

Try the draft.

Choose an approved model, write a prompt, and run a scripted conversation against the active rubric. Bench tools are off. Save or export the configuration; deployment is a separate step.

Questions,
answered.

Where are the controls in the app?

Quality contains AI performance and handoffs. Approvals contains held plans, decisions, and matches. Settings contains decision rules, automations, guardrails, skills, and draft agents.

Does a successful bench test deploy an agent?

No. The bench tests a prompt with tools off and scores the transcript where a rubric is available. Deployment and connected-tool checks happen separately.

What can the action gateway change today?

Supported gateway actions acknowledge, resolve, or reopen internal BearScope worklist records. They do not send customer messages or change fulfillment in another system.

Can we start in shadow?

Yes, for configured decision panels. Shadow results record what was assessed without executing an action. The app shows the mode in use.

Make it part of your team.

Book a demo