Creating and Managing Teammates
6 minute read
Why Build Custom Teammates
The AI Team arrives with a roster of specialized teammates that cover core disciplines such as SRE, security, and cost management. Custom teammates extend that foundation when you need an assistant steeped in your own terminology, tooling, or approval paths. Unlike specialists, they begin with no connectors, tone, or permission—every capability is something you design.
Think of the split this way:
- Specialized teammates provide broad domain coverage on day one and stay maintained by Edge Delta.
- Custom teammates let you encode company-specific knowledge, integrate niche systems, and automate rituals that only your team performs.
The sections below walk through planning, building, and operating these bespoke assistants without repeating the onboarding work covered in the Getting Started guide.
Addressing the Prompt Engineering Challenge
Creating effective custom teammates traditionally requires deep expertise in prompt engineering—crafting system prompts that precisely define responsibilities, tool usage patterns, decision boundaries, and communication style. Organizations without this expertise face a difficult choice: invest significant time learning these patterns or accept suboptimal agent behavior.
Edge Delta addresses this barrier through an AI-powered teammate builder embedded directly in the creation workflow. Instead of writing system prompts from scratch, you describe what you want the teammate to accomplish in natural language: “I need someone to monitor Kubernetes pod health and alert me to issues” or “Create an agent responsible for writing quality product research documents focusing on purity, equality, and competitive analysis.”
As you describe requirements through conversational interaction, the builder:
- Generates production-ready system prompts that encode your requirements into structured instructions the foundation model understands
- Suggests appropriate connectors based on the tools and data sources your description implies
- Proposes periodic task schedules when your requirements suggest automated monitoring or reporting
- Recommends permission boundaries based on the risk profile of proposed operations
The builder updates configuration in real-time as the conversation progresses, allowing you to review generated prompts, adjust suggested connectors, and refine behavior iteratively. This approach bridges the expertise gap: teams without prompt engineering knowledge gain reliable custom teammates, while experts use the builder as a starting point that accelerates their workflow.
The platform also implements guardrails that prevent teammates from operating outside intended domains. Attempts to create teammates for use cases beyond current platform capabilities (sales automation, financial analysis without appropriate connectors) receive clear guidance about limitations and alternative approaches. These guardrails evolve as the connector ecosystem expands to support broader operational domains.
Design the Persona
Lasting teammates start with a crisp problem statement. Capture three notes before you open the builder:
- Audience and scope – Who will ask for help, and which questions should the teammate refuse?
- Data sources and tools – Which connectors expose the telemetry or systems the teammate needs?
- Decision boundaries – Can it draft fixes, execute changes with approval, or only analyze and advise?
A short design brief keeps the build session focused and prevents the assistant from ballooning into a generalist.
Build Workflow
Launch AI Team → Teammates → Add teammate. The AI-assisted form uses your natural-language description to propose prompts, connectors, and periodic tasks. Use that as a draft, then fine-tune each element.
Step-by-Step Creation
- Navigate to the Teammates tab

Click + Add Teammate
You’ll see an AI-powered form builder. Type what you want the teammate to do (e.g., “I need someone to monitor Kubernetes pod health and alert me to issues”)

As you discuss your requirements, the AI updates the Configure tab in real-time. You can review and adjust the generated configuration before saving.
Configuration Checklist
Follow this checklist as you work through the builder:
- Identity – Choose a clear name, avatar, and one-line description so teammates immediately understand the persona.
- Describe the teammate – Use the AI chat pane to describe what you want the teammate to do. The AI will generate the system prompt automatically. Optionally review and tweak the generated prompt if needed, though this is rarely necessary.
- Connectors and tools – Assign the minimum set required. Mark risky operations (deploy, delete, change configuration) as “Ask permission” so the AI requests approval before acting.
- Channels – Limit availability to the places where insight is useful (for example
#incident-response,#finance-ops, or direct messages). Tenant-wide access is best reserved for broadly applicable personas. - Periodic tasks – Schedule automated reports or checks only when someone will act on the output. Define the task, cadence, and destination channel, then review the first few runs for signal quality.
Ensure the appropriate connectors are assigned to your custom teammate in the Connectors section, then click Save to create your teammate.
Your new teammate appears in the Teammates list and is available for direct messages. Invite the teammate to relevant channels where they should participate.
Note: Built-in specialized teammates automatically receive connector assignments when you configure connectors. Custom teammates require manual connector assignment.
Managing Teammates
Edit a Teammate
- Go to the Teammates tab
- Click on the teammate card
- Click Edit or the settings icon
- Modify configuration, system prompt, or connectors
- Click Save
Toggle Teammate Status
Each teammate card shows an Online/Offline toggle. Disable a teammate temporarily without deleting it.
Clone a Teammate
Duplicate a production-ready persona to create a staging or regional version with adjusted access.
Review and Update Regularly
Revisit prompts, connectors, and permissions after architecture changes. Retire unused teammates to keep the roster tidy. Document major changes so other administrators understand current behavior.
Best Practices
- Start narrow: Launch with a specific workflow, then broaden once the teammate proves reliable.
- Limit permissions: Grant only the connectors and tools required today; revisit access quarterly.
- Keep prompts current: Update instructions when processes change and include examples for tricky tasks.
- Curate periodic tasks: Disable or adjust any automated report that stops driving action.
- Share expectations: Document the teammate’s purpose and limitations in the description so colleagues engage it appropriately.
Troubleshooting
- Generic answers – Refine the system prompt with concrete tasks, decision criteria, and examples.
- Missing data – Confirm the necessary connectors exist, credentials are valid, and the teammate is assigned to them.
- Overly verbose responses – Add formatting guidance (“Limit summaries to three bullets”) to the system prompt.
- Boundary issues – Strengthen the rules section and ensure risky actions require approval or are disabled.
- Noisy periodic tasks – Raise thresholds, reduce cadence, or move the report to a smaller audience.
Related Resources
- AI Team Overview — understand how teammates fit into the broader orchestration flow.
- Getting Started with AI Team — review channel setup, permissions, and first-run considerations.
- Specialized Teammates — see the capabilities you can pair with or extend.
- AI Team Support — reach out when you need tailored assistance.