JumpinoI build integrations around the existing workflow and the data the assistant is allowed to use. We define when it should answer, when it should ask for clarification and when it should hand the request to a person. Sending messages or changing records needs explicit permission rules.
Wegweiser Leben includes a custom assistant alongside its Astro website and Rust backend. For your project, I start with representative questions and expected answers. We test unsupported requests and failure cases, then review operating costs and response quality before expanding access.
What we can build
- An assistant connected to the information you approve
- Defined boundaries for data access and tool actions
- A set of realistic evaluation questions and failure cases
- Human review, error handling and usage monitoring
When it makes sense
Useful when the same information is repeatedly requested or a manual process has clear reviewable steps. A simple form, rule or search feature may solve the problem more reliably than AI.
Know what happens before we start
Agree the scope
The goal, pages, features, languages and materials become a proposal with costs, timing and responsibilities.
Review an early version
You see the structure and workflows before release. We discuss extra requests separately so the scope stays clear.
Check and hand over
We check phones and desktops, forms and access. Handover covers the agreed repositories, credentials, instructions and maintenance arrangements.
A handover you can use
- Agreed code and accounts, with credentials shared through an appropriate channel.
- Instructions for content updates and recurring operations.
- Documented external costs, backups and operational responsibilities.
After the launch
Updates, backup checks and new features need a plan. We agree what is included, who responds and how requests are handled. Round-the-clock support is not included unless explicitly agreed.
Hosting and maintenance →Questions before starting
Can an AI assistant answer incorrectly?
Yes. I design it to use approved sources, acknowledge missing information and hand off uncertain requests. Evaluation and human review reduce risk but do not make every answer correct.
Will company data be sent to an AI provider?
That depends on the chosen model and hosting. Before implementation, we agree which data may leave your systems and check the provider settings and retention terms.