Agents
Give an agent a role, standards, tools, and knowledge so it can help with related work wherever a request starts.
Give related work to the same agent
Set up an agent to help with an ongoing responsibility, such as customer relationships, recruiting, or operations. Give it standards, useful methods, and access to the tools it needs. Then bring it different jobs within that role.
An agent is a saved teammate with a role and a way of working. A thread is where it carries out a particular job. You can ask the same agent to prepare for a meeting today, draft a follow-up tomorrow, and review an account before renewal next month.
One role, several jobs
This agent helps you stay prepared and follow through with customers. It can learn a separate process for each job.
The agent's System Prompt sets its role and standards across those jobs. Skills hold the steps and checks for particular jobs. A new request supplies the customer, date, and sources for the work at hand. You do not need a new agent for every brief or follow-up.
Create another agent when the responsibility needs different standing instructions, access, or oversight. If two kinds of work pull one agent toward conflicting behavior, separate the roles. See configuration for what an agent keeps in its setup.
What an agent brings to the work
Role and standards
Its System Prompt describes what it is responsible for, how it should work with you, and when to ask. Those directions apply across its jobs.
Process
Skills give the agent procedures and checks it can use when a request calls for them. Scripts in a skill can handle work with exact rules.
Tools and connections
Built-in tools and connected accounts let the agent research, work with files, and use services it has permission to access.
Knowledge
Memories hold lasting facts and preferences. Documents provide larger source material. Knowledge access controls what the agent can find and use.
Model and limits
Model, reasoning, and run settings shape how the agent works within its available resources.
Autonomy
Permissions and approval settings determine when the agent can act on its own and when it needs you to review a change.
The agent keeps these settings and resources across jobs. A thread holds the request and working material for one job. Skills provide a process the agent can follow when that job calls for it.
Having a connection or a document available does not mean the agent uses it in every answer. The request, its permissions, and its thread context determine what it can work with at that moment.
Where work starts
You can message an agent directly, or a schedule, webhook, or connected channel can start work for it. Each request becomes work in a thread, where you can follow the agent's progress, answer questions, and review the result. You can find past runs in the agent's Activity. The result can also be delivered elsewhere.
Your message
Give the agent a job, add its sources, and continue the work in the same thread when the job needs another pass.
A schedule
Have the agent start one of its jobs at a regular time, such as preparing a brief before the day's calls.
Another entry point
Let Slack, Telegram, email, a webhook, or another agent send work to the same role where its permissions allow it.
Govern the work your agent can do
An agent can help with several jobs while keeping clear limits on what it can access and change. Set those limits alongside its role, then review how they work in real runs.
Choose its access
Connect the accounts and enable the tools the role needs. Instructions describe the job; they do not grant access to an account or resource.
Set action boundaries
Decide which actions can proceed and which need approval. Review these controls in your agent’s configuration.
Review what happened
Follow work in its thread and inspect past runs in Activity. Review requests for approval and test the actual entry point before relying on unattended work.
Delegation is another choice to review: decide which other agents this agent can ask for help. See agent delegation for its access and approval boundaries.
What carries forward
The agent keeps its saved role, access, and methods. Each thread keeps the conversation and working material for its job. A new thread does not automatically receive every detail from an earlier one. Save reusable methods as skills and lasting facts as memories so they can help future work when the agent can access them. See Thread context for the boundaries.
Feedback from a thread can lead to a proposed skill, memory, or instruction update. Review what the agent proposes before treating it as part of its future setup. Teach your agent how you work shows how to turn a result you trust into a reusable method.
Create or reuse an agent
You can start with work you have already improved in a thread, then ask Hyperagent to propose an agent for the broader responsibility. Review its role and save the processes it needs.
Your team may already have an agent that fits. Workspace members can work with agents their team makes available. Where Add to Personal is offered, you can use a workspace agent in Personal with your own accounts and settings.