> ## Documentation Index > Fetch the complete documentation index at: https://aidocs.ethanbragdon.icu/llms.txt > Use this file to discover all available pages before exploring further. # Create a Custom Agent > Create, scope, test, and save a reusable Starfire AI agent with durable instructions and appropriate capabilities. # Create a custom agent Create an agent when you expect to reuse the same role, behavior, or workflow framing across many tasks. ## Create the agent 1. Open the Agents area in Starfire AI. 2. Create a new agent. 3. Choose a clear name and description. 4. Write durable instructions that define role, priorities, boundaries, evidence handling, and output expectations. 5. Select model/tool capabilities when those controls are available. 6. Attach Project or Knowledge context only when it is broadly relevant to the agent's role. 7. Save the agent. 8. Test it before using it in automation. ## Name by responsibility Prefer names that tell users what the agent does. Good examples: * `TypeScript Code Reviewer` * `Release Manager` * `Support Triage` * `Documentation Writer` ## Keep the role coherent One agent should not try to be a code reviewer, marketing writer, billing administrator, and security responder at the same time. Separate roles make instructions easier to maintain and permissions easier to reason about. ## Choose context intentionally Do not attach every Knowledge source or project to every agent. Extra context can introduce irrelevant information and unnecessarily broaden access. ## Test before production use Test: * a normal task * an ambiguous task * a task with missing information * a request that conflicts with the agent's rules * a task requiring a tool the agent does not have A reliable agent should clearly expose limitations instead of claiming unavailable work was completed. Design the durable behavioral contract. Evaluate behavior before automation.