Give your internal AI agents a shared understanding of how your team works. Hubzoid brings your knowledge, rules and tools together in an open-source platform for chat, workflows and connected AI assistants.
What is an internal AI agent?
An internal AI agent is an AI agent that works for your team. It helps finance check invoices, engineering review changes or support prepare replies, using your team's knowledge and the tools you allow.
Unlike a customer-facing bot, an internal agent serves the people inside your organisation. Chatbots, agents and workflows can overlap, but describe different things:
| Term | What it means | Example |
|---|---|---|
| Chatbot | A conversation where someone asks and gets an answer | “What is our travel policy?” |
| Agent | Works toward a goal, choosing steps and allowed tools | “Check these invoices against our orders.” |
| Workflow | A defined sequence, run on demand or on a schedule | “Prepare the sales report every Monday.” |
An agent can work through chat or inside a workflow. The useful part is giving it a shared understanding of how your team works.
Start with what your team knows
Your policies, processes and past decisions are the context a model does not know by default. Bring the useful parts together, so every new agent does not start from scratch.
- What we knowInternal knowledge
Product details, policies, team terminology.
- How we workRules and skills
Review checklists, procedures, good examples.
- What we can useApproved tools
Connections to the systems the work depends on.
One brain. Different agents around it.
In Hubzoid, that shared foundation is the Hub. Each agent draws on the relevant knowledge and skills, then follows instructions for its own job.
Knowledge
Rules · Skills
Authorized tools
- PeoplePolicy assistant
Explain the team's policies.
- FinanceInvoice checker
Flag differences against orders.
- EngineeringReview agent
Check changes against team standards.
- SupportReply drafter
Prepare answers from product knowledge.
Maintain a shared rule in the Hub, instead of copying it into every agent. A Hub can serve one team, a product or a company. It does not have to hold everything your organisation knows.
Different jobs. Useful work for your team.
Start with work someone already repeats. Give the agent trusted sources and a clear result to prepare.
- 01 / Answer“Can I expense this trip?”
Travel policy → policy assistant → an answer with the source.
- 02 / Check“Does this invoice match our order?”
Order and invoice → invoice checker → differences for finance to review.
- 03 / Prepare“Help me reply to this customer.”
Product notes → reply drafter → a draft for your support teammate.
- 04 / Report“What needs attention this week?”
Team data and reporting rules → weekly reporter → a summary for the team lead.
The support agent helps your teammate serve a customer. That is still internal work: your teammate reviews and sends the reply. See five practical use cases for more internal agent jobs.
One engineering request, three agents
Imagine a request to add invoice exports. A planning agent helps clarify the requirements. A coding agent prepares a draft pull request. A review agent checks the changes against the requirements and your team's standards.
- Clarify requirements
- Build a draft PR
- Review the PR
Each agent has a different job, but draws on the same foundation. The requirements and draft pass between steps, so the reviewer can check what was asked for. Software Factory puts this pattern to work.
Meet your team where the work happens
The same Hub can support a conversation, a recurring workflow or a supported AI assistant connected through MCP.
- Chat
Ask a question.
Request some work. - Workflows
Run a repeatable task.
Schedule a report. - Your AI assistant
Connect through MCP.
Reuse the Hub's context.
Shared context does not mean identical conversations or permissions. Use identity and access controls to give people only the access they need, require review before consequential actions, and name an owner to keep the knowledge current.
How to build an internal AI agent
Start with one useful job, then reuse its foundation for the next agent.
- Define the job. Pick a recurring task, name the person who needs the result and describe what a good answer or draft looks like.
- Gather trusted knowledge. Bring together the relevant policies, procedures and a few examples with known outcomes. Keep sources current and give someone responsibility for maintaining them.
- Set the boundaries. Decide which tools the agent may use, who can use them and what needs approval. Start with reading and drafting before allowing changes to other systems.
- Test known and awkward cases. Include missing documents, conflicting rules and requests without permission. Use evals to check whether later changes make results worse.
- Share and maintain it. Let a small group try it, review the results and correct the shared knowledge. Add schedules or more agents once the first job is useful.
The quickstart walks through your first Hub. You can also start from a ready-to-use agent or bring an agent you already use.
Common questions about internal AI agents
Does an internal agent have to be self-hosted?
No. Internal describes who it serves, not where it runs. It can run on a laptop, your server or a hosted service. Hubzoid is open source under the Apache 2.0 license and provides a self-hosted route.
Do I need to train a model?
Usually, no. Start with an existing model and supply relevant knowledge, instructions and tools when it works. Keeping that context accurate is still your team's responsibility.
Can agents share knowledge without sharing access?
Yes. A shared Hub does not mean everyone can use every tool. Hubzoid checks restricted tools against the person's permissions. Scope sensitive knowledge appropriately too, rather than assuming tool permissions protect every document.
When is a simple workflow enough?
If every step is fixed, ordinary automation may be enough. Use an agent when the task needs interpretation, such as reading a request, comparing evidence or preparing a response within your rules.
Put your team's knowledge to work
Start small: one job, trusted context and a person who checks the result. Grow the shared brain as your team finds more useful work for it. If you want help building it, build with us.