What is missing is context
Where you are now
An AI assistant is in use. Each employee uses it alone: explains to the agent what the company is, pastes in data, gets a generic answer and checks it themselves. Next time, from scratch again. Skills are not centralized and are underused.
What is missing
Context. The agent does not know how the company works: which system is for what, what a process looks like, what a metric means, who is allowed to see what. Without that, every answer is generic.
What we bring
GRIM: the company's context and the connection to its systems, via MCP. Experts manage and maintain it with your teams. Employee training uses practical AI tasks and tools, by department.
What changes for an employee
Today
Opens the assistant, explains what the company is, pastes in data, gets a generic answer, checks it alone. Next time, from scratch again. Information and results are not managed.
With GRIM
Asks the question. The agent takes the data from the systems, applies the procedure the team has already approved and returns a draft with its source. What one team learns applies to everyone.
What MCP is
MCP (Model Context Protocol) provides a structured way for an AI agent to access a company's systems and knowledge. The way USB is the standard for devices, MCP is the standard for AI: one server, on the company's side, gives the assistant data, documents and procedures, and checks who is allowed what.
01
Employees
Claude, GPT, the tools they already use
- Ask in their own words, in the tool they already use.
- Get an answer, a draft or a report, with the data source.
02
GRIM
the company's MCP server
- Context: documents, skills, queries, metrics.
- Permissions: who may see what, checked on every call.
- One server for Claude and GPT.
03
Company systems
data stays where it is
- CRM, ERP, HLiD, data warehouse, ticketing, Teams, web and marketing.
Without MCP
- Employees copy data into the chat by hand, every time.
- The agent knows neither the systems nor the processes; answers are generic.
- Every connection to a system is a separate project, for every tool separately.
With GRIM
- The agent asks the server and gets the data from the systems at that moment.
- Along with the data it gets the procedure the team approved (a skill), so it works the way you work.
- One server for Claude and GPT. Permissions are checked on every call; nothing is copied into the model.
MCP is an open standard supported by Anthropic (Claude) and OpenAI (GPT). Nibelung builds and maintains GRIM.
How the agent gets context
GRIM holds the company's context: documents, procedures, queries, metrics and permissions. The agent receives it through MCP when it needs it. Employees work in Claude; GRIM gives it what it has to know, in the background.
Stories
Context in four parts. Documents: instructions, system descriptions, projects. Skills: written procedures, for example how open invoices are looked up. Queries and Metrics: queries and metrics from databases and applications, with accuracy checks.
Apps & systems
All systems, applications and databases in one place, with the Stories that refer to them. An employee sees how something is done; the agent gets the same through MCP.
Boards
Views and modules built on Stories data: reports, automations, alerts. Six starting boards, custom ones as needed.
People
Permissions by entity, department and role. Nobody sees data they have no access to, not even through the agent.
The platform
- Works in the channels you already use: web, Teams, WhatsApp, e-mail, phone.
- Metrics with an owner, a formula and a source, not free-hand calculation.
- Data per entity: several companies and countries in one system.
- A separate installation per customer; data stays in the EU.
- Claude and GPT alike, with no separate integration per tool.
Forward deployed engineers
Context, skills, connectors and permissions do not create or maintain themselves. That is the work of forward deployed engineers: our experts, who sit in your teams rather than in a vendor's office. The system is not trained, it is set up; training is for employees only.
01
A task from the team
a ticket, a report, a campaign
02
Context and skill
the expert and the team build them on that task
03
Result in the assistant
a draft for review, with the data source
04
Stays in GRIM
the whole team uses it from then on
The next task starts from an existing skill, not from zero.
What they do
- Write and test skills with the team, on real tasks.
- Define queries and metrics and check them with the data owners.
- Build and maintain connectors as systems change.
- Set permissions by entity, department and role.
- Check answer quality and correct the context.
- Run training and labs per department.
What it means for an employee
No explaining the company to the agent, no copying data. Ask, get an answer from real data, by the procedure the team approved.
What one team learns applies to everyone. A new colleague starts from the same skills and documents.
Ready to give your agents context?
A first conversation covers your systems, your teams and the first three tasks worth a skill.