Agiled Docs

Agent settings control AI-related workspace behavior where available.

Treat agent settings like access-control settings. They decide who can use AI surfaces and what kinds of workspace actions can be attempted.

Workspace Assistant

The settings page shows the workspace assistant configuration, including:

Runtime model.
Server-selected tools.
Approval requirements for gated write actions.

Use Open assistant to move from settings to the assistant surface.

Before changing assistant settings, confirm which users rely on the assistant and which workflows are currently active. A broad settings change can affect chat, AI workers, and tool-assisted operational review.

Review Access

Review who can use agents, which tools are available, and how AI workers can access workspace context. Keep access limited to the people and workflows that need it.

Use the same review discipline you use for roles and integrations. A user who can run an agent with broad context may be able to summarize sensitive records even when they are not editing them directly. A worker with write tools may change records, send messages, or create follow-up work depending on its configuration.

Tool And Context Review

Review agent tools and context sources together. A tool may be safe when limited to one module but risky when paired with broad CRM, finance, HRM, or file context.

For each enabled tool, decide:

what records it can read
what records it can change
who reviews the output
whether approval is required
how to pause the workflow if results look wrong

Approval Gates

Approval gates protect sensitive write actions. Keep them enabled for actions that can change records, send messages, trigger automation, or expose workspace data.

Before lowering approval requirements, test the worker or assistant flow with a single low-risk record. Review the prompt, context sources, allowed tools, and resulting action. If the output still needs human judgment, keep approval on and document who should review the action.

Review Regularly

Review agent settings after adding new AI workers, enabling new tools, changing roles, or connecting integrations. If a worker no longer needs a tool or context source, remove it.

Also review settings after a support incident, data import, permission change, or connected-app change. Agent behavior can change when new records or tools become available, even if the agent instructions were not edited.

Change Management

When enabling new AI capabilities, start with read-oriented use cases and one low-risk record. Document who reviews outputs, which actions require approval, and when a worker should be paused. Expand access only after the team has seen successful runs and understands the review process.

If an AI surface returns unexpected output, inspect the worker or assistant context, allowed tools, recent settings changes, and related permissions before changing prompts broadly.

Incident Response

If an agent or worker acts unexpectedly, pause the worker or remove the risky tool first. Then review run history, prompts, context sources, permissions, and recent data changes before re-enabling it.

Audit Before Expanding Access

Before giving more users or workers access to AI tools, review a recent set of runs. Look for unsupported assumptions, missing context, unnecessary tool calls, and actions that still needed human judgment. Expand access only when the current workflow is understandable and reviewable.

For finance, HRM, files, and customer communication, keep a named human owner for approvals. AI settings should make the reviewer clear; they should not make responsibility ambiguous.

Agent Settings