A prompt belongs to one person. An agent belongs to the whole organization.
An agent captures what makes a good request: instructions, knowledge, tools, and model. Once tested and shared, it gives everyone on the team the same result, backed by sources.
Everyone phrases their AI request differently. So everyone gets a different result.
One person writes three lines, another half a page; one attaches the policy, the other doesn't. An agent ends this: it works for every person to the same specification, with the same knowledge and the same permissions.
Define
Creating an agent takes four inputs. Instructions, knowledge, tools, and model.
Your team sets these four inputs once, plus an input form if needed. Anyone who uses the agent afterward only writes what they want checked.
You describe what the agent checks, how it answers, and in what format. For the Vertragsprüfer AVV, that is a note with a citation and a supplementary text for each gap.
Test
Test new versions first, then approve them. In the Playground, with real documents.
A new version of the agent first runs in the Playground against real contracts, version 3 and version 4 side by side. Only when the results are right does the new version apply to everyone who uses the agent.
- Old and new side by sideThe live comparison shows the same input in both versions. You see deviations before anyone else does.
- Every version can be restoredThe version history records what changed and when. An earlier version is active again in one click.
- Tested with your contractsTests use contracts whose review outcome your organization already knows. That way, every deviation shows a real difference.
Share
Share an agent with people, groups, or the whole organization. Maintained in one place, used by everyone.
Anyone who uses a shared agent gets the same instructions, the same knowledge, and the same permissions. Only the person who maintains it can change it.
Version 4 goes to procurement, data protection, and IT security. From today, all three review with the same version, no one with a copy of their own.
- Procurement: before signingProcurement uploads a new supplier's DPA to the form and has the note with citation before signing.
- Data protection: right in the chatThe data protection officer brings the Vertragsprüfer AVV into her chat on a third-country transfer with “/”; the context stays in the same conversation.
- Organization: via the MarketplaceIf the agent proves itself, you make it available to everyone via the Marketplace. Every new department starts with version 4.
Example · Agent chains
Several agents in a row in one chat. Called with “/”, each with its own permissions.
With “/”, you bring any agent available to you into the ongoing conversation. An example: the Vertragsprüfer AVV reviews, the Datenschutz Navigator assesses the third-country transfer, the Corporate Mail Writer drafts the response. Nothing is sent until procurement agrees.
Industrie AG
Task: /Vertragsprüfer AVV reviews the Spedition Keller DPA, then /Datenschutz Navigator and /Corporate Mail Writer
/Vertragsprüfer AVV has reviewed § 6
/Datenschutz Navigator has assessed the transfer
/Corporate Mail Writer has drafted the response
Email to Spedition Keller
Spedition Keller receives the response after procurement approves it.
Each agent uses only its own tools: read SharePoint and Confluence, write the email only after the OK.
Agent Library
Several hundred pretrained agents. Adopt them and adapt them to your requirements.
From the Vertragsprüfer AVV to the Onboarding Coach: several hundred pretrained agents in ten industries. You adopt one and adapt its instructions and knowledge to your requirements.
Limits
An agent may only do what its task requires.
Admins decide once for the organisation which locations, models and tools are enabled. Your team then decides per agent what it reads, what it prepares and what it never does alone.
FAQ
When an agent pays off and who builds it.
As soon as the same request comes up again and everyone should get the same result. The agent captures instructions, knowledge, tools, and model. If it should run regularly without a prompt, it belongs in an automation.
Yes. In chat, you call agents one after another with “/”; each works with its own instructions, knowledge, and tools, and the result stays in the same conversation. If a chain should run regularly, save it as an orchestration with a schedule.
Yes, but never alone. Reading steps run freely; emails, meetings and data changes wait for a person's OK. You set which tools, data rooms and recipients an agent uses for each agent. What is allowed at all is decided by your admins for the whole company.
Your business unit builds it, without programming: write instructions, assign knowledge, enable tools, test in the Playground, share. It is faster with an agent from the Agent Library that you adapt to your requirements. Our Customer Success team supports the first agents during onboarding.
No. In SharePoint, the permissions of the person using the agent apply; in addition, you define per agent which tools, data spaces, and recipients are allowed.
No. An agent works when someone calls it, in chat with “/” or via its form. A trigger such as a new file or a meeting comes from an automation.
The tools your admins enable for the company, such as SharePoint, Outlook, Confluence or Jira. Which systems are ready to connect is listed under Integrations; we connect others individually as an additional service.
Yes. Every tool call appears in the history with arguments, response, and duration: reproducible, exportable, and auditable.
Your most frequent request. Your first agent.
During onboarding, our Customer Success team supports your first agent: on your documents, tested in the Playground, shared with the departments that need it.
After your request
- 01You name the request that comes back every week
- 02You see in the neuland.ai HUB how an agent from the library handles such a request
- 03You decide which departments it starts with