Skip to content

Agent Settings

Every agent has a detail page where you can configure exactly how it behaves, what it can access, and who it works with. To open it, click on any agent card from the Agents page. The header shows the agent’s name, a switch that suspends or reactivates the agent, and an Inbox button that opens the agent’s email inbox. Owners and admins also get a Rename agent control (hover over the name on desktop) and a Publish to Marketplace button, which becomes Manage Listing once the agent is published. This guide walks through each section of the agent detail page from top to bottom.

The persona is the core instruction set that shapes your agent’s identity and behaviour. Click Edit to write (or refine) the instructions your agent follows in every conversation.

Your persona should cover:

  • Who the agent is and what it does
  • The tone and communication style to use
  • What the agent should and should not do
  • When to escalate or hand off to another agent or a human

You can update the persona at any time. Changes take effect immediately, including in conversations that are already open.

Company Directives are organisation-wide rules that apply to every agent automatically. They appear on the agent detail page as a read-only section so you can see what global rules are in effect, but you cannot edit them here.

Examples of company directives might include:

  • “Always greet customers by name when available.”
  • “Never share internal pricing margins with external users.”
  • “Include a disclaimer when providing legal or financial information.”

Company directives are managed from the Settings area of your organisation. They ensure consistency across all agents without having to repeat instructions in every persona. See Directives for the full picture.

Unlike company directives, agent directives are specific to this individual agent. Click Edit to add a directive, switch one on or off, change its priority or remove it. To change a directive’s wording, remove it and add it again. Each directive has a category (Core Values, Can Do, Cannot Do, System) and a priority (High, Medium, Low) that controls how strictly the agent treats it.

Use agent directives to add targeted rules without cluttering the persona. For example:

  • “When a customer asks about warranty, always reference the warranty policy document.”
  • “If the conversation involves a job number, look it up in the job system before responding.”

Directives are applied alongside the persona and company directives, giving you a layered approach to controlling agent behaviour. See Directives for details on how the levels combine.

These are integrations configured at the organisation level that are available to all agents. They appear as a read-only list so you can see what is connected — they are managed from the Integration Hub. For example, a helpdesk or accounting integration your organisation has connected in the Integration Hub.

These are integrations enabled specifically for this agent. You can toggle which integrations the agent has access to, or configure new ones that only this agent will use.

For example, you might connect a helpdesk integration only to your “Customer Support” agent, while keeping your accounting integration available to all agents.

See Tools Reference for a full list of integration tools available to agents.

The Schedules section lists the recurring and one-off tasks this agent runs automatically — for example:

  • “Every Monday at 8am, check for overdue invoices and send a summary to the finance channel.”
  • “Every day at 5pm, compile a report of today’s completed jobs.”

Each entry shows when it runs (in plain language), when it runs next and how its last run went. Owners and admins can switch a schedule off or on, or delete it. Schedules are created by asking the agent in chat — see Schedules & Tasks for the full guide.

Task History shows the agent’s recent background task executions — scheduled tasks and other work that ran outside a live conversation. Each entry shows the task’s name, type, status (queued, running, completed, or failed), when it started, how long it took and how many tool calls it made. Owners and admins can open Details to see the result or error. Use this to confirm scheduled tasks are running as expected.

The Configuration section covers the agent’s core operating settings:

  • AI model — Choose the model this agent runs on from the dropdown and click Save: Auto (Sprigr’s recommended model, currently Sonnet, for every message), Opus (most capable), Sonnet (balanced), or Haiku (fastest). Extra models can appear if they have been enabled for your organisation. The performance tier chosen when the agent is created sets the initial value; you can change it here at any time and the change applies to open conversations straight away.

  • Role — The agent’s access level, which you can change here:

    • Member — Standard access. The agent can use its assigned tools, search knowledge bases, and communicate with connected agents. Suitable for the vast majority of agents.
    • Owner — Elevated access. The agent skips your approval policies for most actions (it still cannot create agents or change policies on its own) and can read your organisation’s admin and escalation data. Reserve this for orchestrator-style agents that genuinely need it.

    Only owners and admins can give an agent the Owner role. You cannot change the last Owner agent in your organisation back to Member.

Managers are other agents this agent can escalate to. When an agent has managers, it is told to pass complex decisions to one of them when that makes sense. Managers do not monitor the agent or step into its conversations. For example, a “Customer Support Lead” agent could be the manager several frontline support agents escalate to. An agent can have more than one manager.

Owners and admins can add a manager by picking from the list of agents in your organisation.

This section shows which teams this agent belongs to. Teams allow groups of agents to collaborate, delegate tasks, and communicate with each other based on defined rules.

Owners and admins can add the agent to an existing team, or remove it from one, directly from this section. A team added here is joined as a public member.

Assigned users are the people in your organisation who have direct access to this agent. This section is visible to owners and admins. When users are assigned:

  • The agent appears in their sidebar for quick access
  • They can start conversations with the agent directly

Each person you assign gets one of three access levels. The level decides what they can change on this agent:

Access levelWhat they can do
ViewerChat with the agent and read its settings
ManagerEverything a Viewer can, plus change its persona, directives, AI model, other settings and channels; suspend or resume it; and create, edit and run its schedules
AdminEverything a Manager can, plus connect, change and remove its integrations

Some changes stay with your organisation’s owners and admins, whatever a person’s access level: renaming or deleting the agent, changing who has access to it, changing the agent’s own role, and deleting its schedules or turning them on and off.

Access levels only limit team members. Your organisation’s owners and admins can always do everything on every agent.

If someone tries a change their level doesn’t allow, Sprigr tells them which level they need and that an owner or admin can change it. An integration shared by several agents needs Admin access on every one of those agents.

Pick the person, pick their access level, then select Assign. Before anything is saved, Sprigr shows what that level can and can’t do. New assignments start at Viewer unless you pick another level. To change someone’s level, remove them and assign them again with the new level.

Owners and admins can also ask an agent in chat to give someone access. The agent will ask which level you want and confirm what that level allows before it makes the change.

People who were assigned before access levels were introduced were given Admin, so nobody lost access they already had.

Access levels are separate from the Managers section above, which lists the other agents this agent can escalate to.

This is useful for restricting certain agents to specific departments or roles. For example, an HR agent might only be assigned to the management team.

The channels section shows which communication channels this agent is connected to. Channels let your agent receive and respond to messages from external platforms:

  • WebChat — The built-in web chat inside the portal
  • Telegram — Connect your own Telegram bot with its username and token
  • WhatsApp — Available on companion agents only, through Sprigr’s shared WhatsApp number

Slack has no card here: install the Slack app from the App Marketplace instead, as the built-in Slack channel has been retired. Owners and admins can switch channels on and configure them here. For detailed setup instructions, see the Channels guide.

Owners and admins can publish an agent with the Publish to Marketplace button in the page header. Once an agent is linked to a marketplace listing, the header button becomes Manage Listing and a Marketplace section appears with a link to it.

At the bottom of the agent detail page is the Danger Zone, visible to owners and admins. This section contains destructive actions:

  • Delete Agent — Permanently removes the agent. This revokes its API keys, removes all its data, and cannot be undone.

Companion agents have no Danger Zone, because they are tied to their user.

  • Creating Agents — Go back to the basics of creating a new agent if you need a refresher.
  • Agent Teams — Set up teams so your agents can collaborate and delegate tasks.
  • Tools Reference — See the full list of tools your agent can use.
  • Channels — Connect your agent to messaging platforms.
  • Knowledge Bases — Give your agent access to company documents and data.