Your AI agent already helps you prepare a report or answer a recurring question. Now a colleague wants to use it. Sharing the prompt is a start, but it does not share the context, accounts and judgment that made it work for you.
At a glance
- Put the team's instructions, knowledge and procedures in one maintained place.
- Give people access to the capabilities they need, without sharing your credentials.
- Name someone to keep the context current and investigate failures.
Write down what you quietly fix
You know which sales export is current, which branch reports late and what the team means by net sales. When the agent misses something, you correct it before anyone sees the result.
Your colleague does not get that invisible handover. A fluent answer can look correct even when it uses the wrong definition.
Start by recording the details you check every time: the source to trust, the reporting period, the exceptions and who owns the next action. Keep each note short enough for another person to review.
The first thing to share is the knowledge that makes the answer useful.
Share a foundation people can maintain
In Hubzoid, that foundation is the Hub. It brings together written instructions, reference knowledge, reusable skills and tools that reach the systems your team uses.
| Personal setup | Shared team setup |
|---|---|
| Definitions in your notes and head | Maintained notes with an owner |
| Tools using your accounts | Credentials configured for the intended access |
| You notice a bad answer | A reviewer and a place to report corrections |
| You remember to run the report | A schedule, run history and an owner |
The same Hub can support chat, workflows and supported assistants connected through MCP. MCP is the protocol that lets an assistant request context and tools from another system. People can keep their own assistant without maintaining another copy of the team's rules.
Shared knowledge still needs individual access
Two teammates can use the same agent and have different permissions. A store lead might read stock counts while a buyer can also read unit costs.
Hubzoid checks restricted tools in code before they run and records the decision. The model cannot grant itself a permission. The channel matters too: a shared conversation is not an appropriate place to expose information available to only one participant.
Keep private material out of shared notes. Within a Hub, those notes are shared context. Use a separate Hub or an appropriately restricted tool when the audience differs.
Give the agent a way to outlast its builder
Before inviting the team, answer three questions:
- Who maintains the context? Someone must update a changed policy or definition.
- Who manages access? Someone must grant, review and remove permissions.
- Who watches the runs? Someone must notice a failed report and know what to do.
One person can hold all three roles at first. Write the responsibilities down so another person can take over.
Conversation history also needs a boundary. A private chat, a workflow's progress and the team's maintained knowledge serve different purposes. One person's conversation should not silently become everyone's policy.
Start with the result colleagues already ask for
Choose one useful output, such as the weekly stock review. Share it with a small group, keep the first version read-only and collect the questions it cannot answer. Fix the shared context before adding more work.
Teammates using the Hub's web chat do not each need a separate model-provider subscription. The deployment still needs a configured model, hosting and an operator. Connected assistants have their own setup and account requirements.
Next step: read how to bring an existing agent into Hubzoid.
For a short demonstration, watch The Ask agent, episode 6 of Inside AI Agents (opens in a new tab).
