Docs / AI Agents / Custom Agents

Custom Agents

Build your own AI agents with custom system prompts.

The seven built-in agents cover common workflows, but every business has unique needs. The Custom Agent Builder lets you create your own agents with tailored system prompts, so the AI behaves exactly how you need it to for your specific industry, company, or workflow.

Creating a custom agent

Open the Agents panel from the sidebar and click Create Agent. You'll configure four things:

Field Purpose
Name A short label that appears in your agent library (e.g., "HIPAA Compliance Check" or "SaaS Agreement Reviewer").
Description A one-line summary of what the agent does. This helps you (and your team) remember the agent's purpose at a glance.
System prompt The instructions that define how the agent behaves. This is the core of your custom agent — it tells the AI what role to assume, what to focus on, and how to structure its output.
Model Choose which AI model this agent uses. You might use a faster model for simple tasks or a more capable model for nuanced analysis.

Writing effective system prompts

The system prompt is what makes your agent unique. A good system prompt is specific, structured, and tells the AI exactly what to do. Here are some guidelines:

  • Define the role. Start with who the agent is: "You are a healthcare compliance specialist reviewing contracts for HIPAA violations."
  • Specify the task. Be explicit about what the agent should do with the input: "Review the provided contract and identify any clauses that conflict with GDPR requirements."
  • Structure the output. Tell the agent how to format results: "Return your analysis as a numbered list with the clause text, the issue, and your recommended revision."
  • Include domain knowledge. Add specific rules or standards the agent should reference: "Flag any indemnification clauses that exceed 2x the contract value" or "Ensure all data processing terms include a sub-processor notification requirement."

Use cases

Custom agents are useful for any repeatable workflow where you'd otherwise copy-paste the same prompt. Some examples:

  • Industry-specific contract review. A construction company might create an agent that checks subcontractor agreements for lien waiver requirements, insurance minimums, and safety compliance clauses.
  • Company-style email drafting. An agent trained on your company's communication tone and templates, so every outgoing email sounds consistent.
  • Regulatory compliance checks. Agents that flag terms conflicting with GDPR, CCPA, HIPAA, or other regulations relevant to your business.
  • Internal policy enforcement. An agent that checks contracts against your company's standard terms and flags any deviations for legal review.

Managing your agents

Custom agents appear in your agent library alongside the built-in agents. You can edit, duplicate, or delete them at any time. In team workspaces, custom agents are shared with all team members, so an agent created by one person is available to everyone on the team.

Start simple, then refine

You don't need a perfect system prompt on the first try. Create the agent with a basic prompt, run it on a sample input, and iterate. Adjust the instructions based on what the agent gets right and what it misses. The best agents are refined over several iterations.