An agent is an AI specialist that does real work, not just answers questions. Give it a role, a task, and access to your data, and it opens files, runs queries, drafts documents, and hands back finished work. Think of agents as AI employees: you set them up once, and your whole team can put them to work.
Reach for an agent when a task repeats or needs its own data and rules. For one-off questions, plain chat is faster.
An agent bundles four things into one saved setup:
- Instructions: what the agent is for, how it should behave, what tone to use, and what "done" looks like.
- Skills: the skills it reaches for first, such as your pricing rule, your credit scoring, or your brand-voice rewrite.
- Tools and connectors: the actions it can take, such as searching the web, querying a database, or sending an email through a connector.
- Sources: the data it knows about by default, such as your CRM, contract library, or policy folder.
Open an agent and you get a chat that already knows what it is doing. You just tell it the specific thing you need today.
Build an agent by describing the role
The fastest way is to describe the specialist in plain language. Aleria drafts the whole setup for you. You review it, save it, and use it.
1. Describe the role
Open Agents, start a new agent, and write the brief. For example:
"A compliance reviewer for our bank's merchant onboarding. It reads a merchant pack, checks it against our policy, flags any risks, and drafts a summary for the approver. Be cautious and precise, and cite the policy clause by number."
Attach the policy documents, the kind of output you want, and any sources it should always check.
2. Aleria drafts the agent
You get back a ready setup: a name, instructions, recommended skills, recommended sources, and example prompts. Everything is editable.
3. Review and tune
Read the instructions and fix anything that feels off, either directly or by asking Aleria to revise it, for example:
- "Be more direct."
- "Always cite the clause number."
- "Do not approve anything marked tier-1 without escalation."
Add or remove skills, and swap sources, until it matches how your team really works.
4. Save and share
Save the agent. By default it is private to you while you refine it. When it is ready, share it with your workspace so anyone can open it. See Sharing and privacy.
Examples of agents teams build
These are common roles teams set up. You build them by describing what they do; use them as a starting point.
General business
- Legal Reviewer: contract review, clause extraction, risk flags, redlining
- HR Partner: policy questions, onboarding playbooks, job-description drafting
- Finance Analyst: statement analysis, reconciliation, variance reporting
- Procurement Analyst: vendor comparison, RFP drafting, spend analysis
- Compliance Officer: policy mapping, audit preparation, evidence collection
- Sales Partner: account research, proposal drafting, pipeline review
Industry-specific
- Real estate: property valuations, due diligence, lease abstraction
- Financial services: credit memos, know-your-customer packets, portfolio analysis
- Telecom: regulatory filings, churn analysis, tariff comparisons
- Government: RFP responses, constituent questions, policy comparison
Use an agent
Open an agent and it behaves like any chat. Upload files, generate outputs, build apps, or let it run a long task in the background. An agent is a starting point, not a cage: you can change its instructions, add a source, or adjust the tone mid-conversation. The setup gives you a head start, and the conversation does the rest.
How an agent works in the background
Agents do not have to stay in a chat window. When you give one a longer job, it runs on its own and keeps a visible, step-by-step log of everything it does. You can close the tab and come back later. An agent moves through a simple life cycle:
- Draft: you are still building it.
- Published: saved and ready for the team to use.
- Running: actively working on a task in the background.
- Stopped: paused or finished.
You can assign standing work to an agent through Tasks, and you can wire agents into Workflows as steps.
Put an agent on a messaging channel
Where this is enabled, you can connect an agent to a messaging channel such as Slack, Telegram, WhatsApp, or Mattermost, so colleagues can work with it where they already chat. For example, a support agent in a Slack channel can answer policy questions for the whole team. This is an add-on that your administrator turns on. See Plans and feature availability.
Adapt an agent someone else built
Every agent is editable. Copy an existing one and turn it into a specialist for your team, for example a Legal Reviewer that reads your jurisdiction's case law and flags clauses in your house style. Your copy is yours, so changes do not affect the original.
You can change the name and description, the instructions, the skills (including custom ones your team built), the sources, and the example prompts.
Share agents across teams
Once you have tuned an agent to how your team works, share it with the workspace. Anyone in the workspace can open it, and new teammates get the same specialist on day one. If your organization runs several workspaces, an administrator can promote an agent so every workspace inherits it.
How agents fit with the rest of Aleria
An agent is how skills and sources come together as a specialist. That specialist works for you in a chat, runs in the background on a task, or becomes a step inside a workflow when the work becomes routine. One agent, many places it shows up.
Common questions
What is the difference between an agent and a regular chat?
A regular chat starts blank, while an agent is a saved specialist that already knows its role, with set instructions, skills, tools, and sources. When you open an agent you get a chat that already knows what it is doing, so you just tell it the specific thing you need today.
Can an agent keep working after I close the tab?
Yes. When you give an agent a longer job it runs on its own in the background and keeps a visible, step-by-step log of everything it does, so you can close the tab and come back later.
What is the difference between an agent and a task?
An agent is the specialist that does the work, and a task is a single job you assign to that agent through Tasks. You set up an agent once, then hand it tasks to carry out.
How do I share an agent with the rest of my team?
By default a new agent is private to you while you refine it, and when it is ready you share it with your workspace so anyone can open it. If your organization runs several workspaces, an administrator can promote an agent so every workspace inherits it.
Can I edit an agent that a colleague built?
Every agent is editable, but the safe way is to copy it first so your changes do not affect the original. On your own copy you can change the name and description, instructions, skills, sources, and example prompts.