The first two weeks of a new teammate follow a familiar script. They get access to everything "just in case", they are asked to "read the channels", and they spend days interrupting colleagues with questions whose answers are buried somewhere in last quarter's messages. In a small team, every one of those interruptions costs real work.
There is a better pattern: give the newcomer the right access from day one, and an AI agent that knows exactly what they should know and nothing more. Here is how to do it in SweetHive.
Start from access, not from documents
Onboarding usually starts with a list of documents to read. Start instead with a question: which parts of the company does this person need to see? In SweetHive the answer is written in your structure. Your hive contains contexts (departments, projects, clients), and groups decide who sees each context. A new designer might need the Studio context and two client projects. They do not need Administration or the Board.
When you add the newcomer to the right groups, they see those contexts, from that level down, and nothing beside them. That single decision also defines what their agent can ever know, because an agent in SweetHive never sees more than the person who owns it.
Step 1: put them in the right groups
- Decide the contexts they need for their first month. Keep it narrow; you can always widen later.
- Invite them as a member and add them to the matching groups. If you use Aristeo, SweetHive's consultant agent, it can propose the placement and send the invitation, but only after you confirm exactly who goes where.
- Check what they will see. Open each context and look at its active groups and people. If something surprises you, fix the group, not the person.
One distinction matters here. Freelancers and short-term collaborators can join as guests: they are free and see only what is shared with their group. Guests cannot use agents, though. If you want the newcomer to have their own agent, invite them as a member.
Step 2: write a welcome message in each context
Agents work from the messages in their contexts. A short welcome post in each context the newcomer joins does two jobs at once: it orients the person, and it gives their agent a clean, up-to-date summary to draw on. A good welcome message covers:
- What this context is for, in one sentence.
- Who decides what, by name.
- The two or three things currently in progress.
- Where the important files are, with one line on what each contains.
- The words the team uses for decisions and blockers, so search finds them.
Attach the handbook and the key documents too: with Agents Node 0.2.23 or later, the agent can read attached PDFs, images and text files in its scope, up to 20 MB each. Still write the substance in the message itself. A line explaining what the attached handbook covers is what search finds first, and it tells the newcomer why the file matters.
Step 3: let them create their agent
On the dashboard of any context, the newcomer finds Your personal agent and can create an agent that covers that context and the ones inside it. It runs on their computer with the free SweetHive Agents Node, using a local model, or on your company's Agents Server if you have one. Either way, it answers inside their visibility.
This is the part that makes onboarding calmer for everyone. The newcomer can ask the agent anything, as often as they like, without worrying about seeming slow. And you do not have to worry about what the agent might reveal, because it simply cannot reach contexts the person cannot see.
Questions to suggest on day one
Give the newcomer a short list of prompts to try. They teach the person how to use the agent and how the team works:
- "Summarize what this context is about and who the main people are."
- "What happened here in the last two weeks? Group it by topic."
- "List the decisions made this month, with dates and who made them."
- "Which questions are still open, and who usually answers them?"
- "Explain the term we keep using for our review process, using messages from this context."
Encourage them to ask the agent to quote its sources. They will learn where things live much faster than by reading in order.
Step 4: widen access deliberately
After two or three weeks, review. Has the person started working with another team? Add them to that group, and their agent follows automatically, with no separate AI permission to update. Has a temporary project ended? Remove the group, and the agent loses that context at the same moment, because SweetHive checks the owner's visibility on every request.
This also works in reverse when someone changes role or leaves. Access is one decision in one place. There is no forgotten chatbot that still remembers the payroll thread.
What this does not replace
An agent is not a mentor. It cannot tell a newcomer which colleague is best at untangling a client's requests, or why a decision was made if nobody wrote the reason down. Keep the human parts of onboarding: a buddy, a first real task, a short check-in at the end of each week. What the agent removes is the low-value part: hunting for context, re-asking questions, reading months of history to find three relevant messages.
A one-page onboarding plan
- Before day one: choose contexts, prepare welcome messages, invite to the right groups.
- Day one: install the Agents Node or pick the company server, create the agent, try the starter prompts.
- Week one: one real task, questions to the agent first and to people second.
- Week three: review access, widen or narrow by group.
Few, but mighty teams cannot afford a month of ramp-up for every hire. They can afford a clear structure and an agent that respects it.
Plan your next onboarding in SweetHive. Start the 14-day free trial, set up the contexts a new teammate would need, and see what their agent can and cannot see.