Agents
What an agent is
An agent is a named specialist with its own expertise, instructions and set of abilities. You talk to one by putting its name after @ in chat:
@fin-analyst what's the P/E ratio of RELIANCE.NS today?
@excel-builder make a monthly budget tracker with a chart per category
@presales what did Acme agree to on the last call?
Agents come in three kinds:
| Kind | Examples | What they're for |
|---|---|---|
| Employees | @gtm @creator @bdr @presales @delivery @builder | They hold jobs, keep notes over weeks, and answer from their own work. |
| Specialists | @web-researcher @data-analyst @report-gen | Experts for one kind of task. You can call them directly, and workflows use them for individual steps. |
| Your own | @legal-reviewer @hiring-screener | Whatever your business needs. See Make your own agent. |
Find them under Build → Agent Pool → Profiles.
Getting agents
- Set up my AI team on the Jobs page adds the six employees and every specialist their work needs.
- Starter Pack on the Profiles screen adds the ready-made specialists. It skips any you already have.
- New Profile makes one of your own.
Talking to an agent
Name it to be certain who answers: start your message with @name.
Or don't. If your message clearly matches an agent's speciality ("summarize this article", "debug this Python error"), AI Partner picks that agent for you. The reply shows which agent answered.
Employees answer first and work when asked. A question to @presales gets an answer from its client notes. A request like "draft the Acme proposal" starts a run in its job, and the result is posted back to you. See Working with your team.
Name two agents to combine them. The first can read the second's notes and ask it questions:
@delivery is data migration in Acme's scope? check with @presales
Built-in specialists
Research
| Agent | Good for |
|---|---|
@web-researcher | Researching a topic across many sources, with citations |
@fact-checker | Checking claims against sources |
@trend-spotter | What's new and moving in a market or topic |
Data and finance
| Agent | Good for |
|---|---|
@fin-analyst | Markets, company financials, valuation |
@data-analyst | Cleaning, analysing and charting data |
@excel-builder | Spreadsheets with real formulas and formatting |
@bookkeeper | Bookkeeping and accounts |
Writing and planning
| Agent | Good for |
|---|---|
@report-gen | Structured reports and documents |
@summarizer | Executive summaries and TL;DRs |
@tech-writer | Documentation, guides, READMEs |
@task-planner | Breaking a goal into a plan |
@project-manager | Plans, timelines and status |
@prompt-architect | Writing and improving AI prompts |
Building
| Agent | Good for |
|---|---|
@python-dev · @node-dev | Writing and running code |
@debugger | Finding why something is broken, and fixing it |
@qa-tester | Testing and checking work |
@shell-ops · @database-admin | Servers, scripts and databases |
Getting things done in the world
| Agent | Good for |
|---|---|
@publisher | Posting or sending finished content to Slack, X, email, Telegram and more |
@inbox-handler | Working through your email and replying in your voice |
@meeting-coordinator | Finding a time and booking meetings |
@submission-specialist | Filling in and submitting web forms: directories, sign-ups, applications |
@agile-pm | Taking a feature from idea to a ready-to-review pull request |
@ops-runner | Deployments and rollbacks |
@cybersec-pm | Authorised security assessments of systems you own. It asks for your authorisation first. |
Some agents lead a small team of their own. @cybersec-pm and @agile-pm hand parts of the work to their specialists, so you only need to talk to the lead.
Make your own agent
Open Profiles → New Profile. If you'd rather not write it yourself, click Generate profile with AI, describe the agent in a sentence, and it fills in the form for you to adjust.
- 1Name and description
The name you'll type after
@, and a line about what it's for. - 2Instructions
Who the agent is and how it should work, in plain language:
You are a careful contracts reviewer for an Indian services business. Flag risky or ambiguous clauses, suggest safer wording, and always recommend a lawyer for anything binding.
- 3What it can use
Tick the abilities it needs, such as web search, reading your files or making documents. Anything unticked is simply unavailable to it. An agent that only reviews contracts has no reason to send messages, so it can't.
- 4Skills (optional)
Assign the skills it should always use. It then applies them without being told.
- 5Required integrations (optional)
If it only makes sense with, say, Gmail connected, add it here. You're told up front when something's missing, instead of seeing a half-finished run.
- 6Keywords (optional)
Words that should bring this agent in automatically:
contract, NDA, clause, liability.
Changes take effect immediately. Built-in agents can be edited the same way.
Give an agent a job
An agent becomes an employee when it holds a job. Open Jobs, create a job, pick your agent as its owner and add wakes. See Make your own job.
Let an agent work on your computer
Tick Your computer in the agent's abilities, and every mention of that agent runs on your connected computer instead of in the cloud. Use it for agents whose whole purpose is on your machine, such as sorting your downloads or updating a spreadsheet on your disk. Your computer needs to be connected with Screen control turned on.
Tips
- One agent, one purpose. Two focused agents beat one that does everything, and you can always combine them by naming both.
- Put facts in notes, not in the instructions. Prices, clients and proof change, so they belong in an employee's living files. Instructions are for how the agent works.
- Keep abilities tight. An agent that doesn't need to send things shouldn't be able to.