Skip to main content

Custom Agents

Custom Agents (also called Custom GPTs) are AI assistants you build and configure for your MSP. Each agent has its own instructions, knowledge base, personality, and scope — so a "Client Onboarding Assistant" agent behaves completely differently from a "Security Incident Helper" agent.

Agents can be used privately by your team, shared with your whole organization, published to the marketplace, or embedded in Studio-built client portals.

What Makes an Agent​

Every agent is defined by:

FieldDescription
Name and descriptionHow the agent is identified in the library and by Jarvis
InstructionsThe system prompt — up to 50,000 characters. Defines the agent's purpose, behavior, tone, and constraints.
ModelWhich AI model the agent uses (inherits from feature routing unless overridden on Enterprise)
Temperature0–2. Lower = more deterministic. Higher = more creative.
Knowledge baseUploaded files, pasted text, or crawled URLs the agent can reference
CapabilitiesWeb browsing, code interpreter, image generation, file analysis — toggle each on/off
Conversation startersPre-written prompts shown to users when they open the agent for the first time
Product scopesWhich The One products this agent can access data from (psa, crm, defend, etc.)
Enabled toolsWhich action tools the agent can invoke (create ticket, send email, etc.)
Confirmation requiredWhether write actions require explicit user confirmation before executing
VisibilityPrivate (only you), Team (your organization), or Organization-wide

Creating an Agent​

  1. Navigate to Custom GPTs in the left navigation
  2. Click New agent (top right)
  3. Choose a starting point:
    • From scratch — blank instructions
    • From template — choose a system template (see below)
    • Clone existing — copy an agent you already own
  4. Fill in the Name and Short description fields
  5. Write your Instructions — be specific. Include the agent's role, what it should and should not do, how to handle edge cases, and what tone to use.
  6. Click Save
ℹ️The most effective agents have detailed, specific instructions. Vague instructions produce inconsistent results. See the Tips section below for writing guidance.

Adding a Knowledge Base​

Agents can answer questions about your specific documentation, client notes, runbooks, and more. Add knowledge to an agent:

Upload files:

  1. Open the agent and click Knowledge → Add files
  2. Upload PDF, DOCX, TXT, or CSV files (up to 50 MB per file)
  3. The platform extracts and indexes the text automatically

Add URLs to crawl:

  1. Click Knowledge → Add URLs
  2. Enter one or more URLs (public websites or documentation portals)
  3. The platform crawls the content and adds it to the agent's knowledge base
  4. Click Refresh to recrawl and update changed content

Paste text directly:

  1. Click Knowledge → Add text
  2. Paste raw text (runbooks, SOPs, product notes)
  3. Give the source a name so you can identify it later

Search the knowledge base: The agent uses semantic search to find relevant knowledge chunks before each response. You can test the search from the Knowledge → Search tab.

System Templates​

The platform includes built-in system templates to get you started:

TemplatePurpose
Customer Support AgentHandles ticket intake, status updates, and FAQ responses for clients
Sales AssistantAnswers pricing questions, qualifies leads, drafts proposals
Technical TroubleshooterWalks users through common IT issues step by step
Security AdvisorExplains security alerts, recommends remediation steps
QBR Prep AssistantPulls data from PSA and Defend to prepare Quarterly Business Review materials
Onboarding ConciergeGuides new clients through your onboarding checklist

To create an agent from a template:

  1. Click New agent → From template
  2. Select the template
  3. Review and edit the pre-populated instructions
  4. Customize for your MSP (replace placeholder client names, policies, etc.)
  5. Click Save

Configuring Capabilities​

Toggle these capabilities from the agent's Settings tab:

Web browsing — Lets the agent fetch and read live web pages when answering questions. Useful for agents that need current information (e.g., security advisories).

Code interpreter — Lets the agent write and execute code to answer analytical questions or generate reports. Charges apply for compute.

Image generation — Lets the agent generate images from text descriptions. Requires image generation quota.

File analysis — Lets users upload files for the agent to analyze (spreadsheets, logs, PDFs). Files are processed in your tenant and not retained after the session.

Product Scopes​

By default, agents can only respond to text input — they cannot read from your PSA, CRM, or other products. Add scopes to give agents data access:

  1. Open the agent, go to Settings → Product scopes
  2. Toggle on the products you want the agent to access:
    • psa — Read tickets, companies, contacts, service time
    • crm — Read companies, contacts, opportunities, activities
    • defend — Read detections, device status, risk scores
    • rmm — Read device inventory, patch status, monitoring alerts
    • cmdb — Read asset records and relationships
    • books — Read invoices and subscription billing
  3. Click Save
⚠️Scopes follow the logged-in user's permissions. An agent with psa scope can only read tickets the current user is allowed to see. Agents do not elevate permissions.

Require Confirmation for Write Actions​

When an agent has tools that can create tickets, send emails, or modify records, enable Require confirmation to prevent accidental writes:

  • When enabled, the agent shows the proposed action and asks "Should I proceed?" before executing
  • When disabled, the agent executes write actions immediately (useful for automated workflows)

Default: enabled for agents with enabled tools.

Publishing to the Marketplace​

Share your agent with other MSPs:

  1. Open the agent, click Publish to marketplace
  2. Fill in the marketplace listing:
    • Category — Customer Support, Security, Sales, etc.
    • Tags — Keywords that help others find it
    • Price — Free or paid (USD amount per install)
  3. Click Publish

Published agents are reviewed by the platform team before going live. The review process takes 1–3 business days.

Agent Analytics​

Track how your agents are being used:

  1. Open the agent
  2. Click the Analytics tab

You will see:

  • Total conversations and unique users
  • Average rating (if published)
  • Tokens consumed per day
  • Cost per day

For published agents, you also see installation count and rating breakdown.

Tips for Writing Effective Instructions​

Be role-specific. Start with a clear role statement:

"You are an MSP support agent for Acme IT. You help clients troubleshoot issues with their Microsoft 365 and networking environment."

Define what the agent should not do:

"Do not provide pricing information. Direct pricing questions to [email protected]."

Specify format:

"Always format troubleshooting steps as a numbered list. Keep responses under 300 words unless the user asks for more detail."

Handle edge cases:

"If you do not know the answer, say so clearly and offer to escalate to a human technician."

Use conversation starters to show users what the agent can do — first impressions matter.

Visibility and Sharing​

VisibilityWho can see and use it
PrivateOnly you
TeamAll members of your organization on the AI Platform
OrganizationAll users across your org and any sub-tenants

Changing visibility from Team to Organization requires Admin role.