Working with AI Agents¶

Understanding AI Agent Types¶

PrimeThink organizes agents into distinct types, each with different visibility and access rules:

System Agents — Built-in agents provided by PrimeThink. These are maintained by the platform and cannot be modified by users.
Catalog Agents — Curated agents available in the shared catalog. Like system agents, these are read-only for all users.
Private Agents — Agents you create for personal use. Only you (the creator) can view, edit, or delete them.
Group Agents — Agents shared with your group. Group members with the appropriate permissions can manage them.
Task Agents — Agents linked to a specific task. Access follows the task's ownership and type.

The table below summarizes who is allowed to edit or delete each type of agent:

Agent Type	Creator (Owner)	Task Owner	Group Admin	Other Users
System	No	—	No	No
Catalog	No	—	No	No
Private	Yes	—	No	No
Task	Yes	Yes	Only for group tasks	No
Group	Yes	—	With permission	No

Note: Platform super administrators can edit or delete any agent regardless of type or ownership.

You can always edit or delete agents you created, except for system and catalog agents, which are platform-managed.
For task-linked agents, if you own the task the agent is attached to, you can also modify the agent — regardless of the task type.
Group admins can manage group agents and agents linked to group tasks, provided they have the required group permission (e.g., Edit Group AI Agents or Edit Group Tasks).
System and catalog agents are managed exclusively by platform administrators and cannot be modified by regular users.
If a task-linked agent has no associated task (or the task has been deleted), only the agent's creator or a super administrator can manage it.

For a full list of agent-related permissions, see Roles and Permissions.

Each agent in PrimeThink can be configured to suit your specific needs:

System Prompt — Define the agent's personality, instructions, and behavior guidelines.
LLM Selection — Choose which Large Language Model powers the agent (e.g., Gemini, Claude, GPT, Mistral, DeepSeek). See Supported LLMs for available options.
Capabilities — Assign tools and capabilities that determine what actions the agent can perform. See Capabilities for details.
Memory — Configure how the agent retains context across conversations. See Memory for more details.

Agents can be tailored for specific domains by combining the right system prompt, LLM, and capabilities:

Data Analysis — Configure agents with data processing capabilities to analyze spreadsheets, generate charts, and extract insights.
Content Creation — Set up agents with writing-focused instructions for drafting documents, emails, and reports.
Research — Equip agents with web search and information retrieval tools to gather and synthesize information.
Technical Support — Build agents with coding capabilities and domain-specific knowledge for troubleshooting and development tasks.
Project Management — Create agents that can interact with tasks, schedules, and team coordination tools.

Agents in PrimeThink are context-aware, meaning they can:

For persistent context across sessions, see Memory.