Build your own agent: from a default to a custom helper in minutes
Worknano ships with agents for tasks, OKRs, CRM, files and setup. You do not have to stop there. An agent is a record like any other, so you can make your own for the way your team works. This guide shows what an agent is made of, three ways to get one, where to attach it and how to write instructions that work.
What an agent is
Every agent has the same parts:
- Name and description, so people and other agents know what it is for.
- Instructions, the standing text that sets its job, its rules and its tone.
- Tools, the actions it may take, such as reading records, searching content or proposing a change. An agent with no tools only drafts text.
- Model, which sets speed and depth. A fast model suits short checks. A balanced one suits multi-step planning.
- Temperature, lower for steady, repeatable answers, higher for more variety.
- Delegation, a list of other agents it may hand work to.
Agents live in the Agents module. They are configuration, not data from Projects or CRM, so no other model needs a link to them. Worknano is customizable with AI, focused on core functionality, and uses the same UI, UX and database in every module. Agents follow that rule: you edit them in the same way as everything else.
Three ways to get one
- Duplicate a default. Open an agent that is close to what you need, such as Chat Agent or Key Result Validator, duplicate it and change the instructions. This is the fastest route, because the structure already works.
- Start blank. In the Agents module, create a new record and fill in name, instructions, tools, model and temperature.
- Ask the Meta Data Assistant. Describe the helper you want and let Meta Data Assistant draft the agent record for you to review.
Where to attach it
- A page assistant. Pick the agent as the assistant for a page, so the sparkles button or Ctrl/Cmd+J opens it there.
- A workflow effect. Add an agent step to a workflow, so it runs when a record changes or on a schedule.
- A chat mention. In a thread, mention the agent by name, or tick Ask to and choose it.
A checklist for good instructions
- State one job in the first sentence. An agent that does one thing is easier to trust.
- Say what it should read, and what it should leave alone.
- Describe the answer you want: a verdict, a table, a short paragraph, a draft.
- List the rules it must never break, such as "never change data without a preview".
- Tell it what to do when it does not know: ask one short question, or say so.
- Give it only the tools it needs. Fewer tools means fewer surprises.
- Test on a few real records, then adjust the text, not the model first.
Honest limits
An agent can only read what the signed-in user may read, and it can only write through the preview steps Worknano provides. It cannot get around permissions, and it will not remember things you never told it. Agents can be wrong, so keep a person in the loop for decisions that matter. Instructions cannot make a missing field or a missing integration appear. If a result is off, the fix is usually clearer instructions or the right tool, not a bigger model.
Next steps: read how to edit an agent for the exact clicks, AI agents for what they can output, and AI assistants for the page drawer. Good examples to copy: Chat Agent, Workflow Creator, Permission Manager and Objective Validator.