Custom software for demanding projects

AI systems, web and mobile applications, process automation, integrations and 3D solutions. From problem analysis and architecture through deployment to production maintenance.

AI systems

AI agents

Agents that carry out tasks in the tools used every day: they read data, make decisions by agreed rules and act in CRM, ERP and internal systems.

RAG and company knowledge

Answers grounded in company documents: manuals, contracts and technical documentation. With the source cited, no guessing.

Process automation

Repetitive work moved to a system: document flow, data extraction, reports and notifications. AI where it helps, plain code where it's enough.

Apps, backend and 3D

Web & mobile applications

Complete applications: frontend, backend, mobile and infrastructure. From architecture to production maintenance.

Backend and integrations

APIs, job queues, databases and connections between systems that today exchange data by hand or through spreadsheets.

3D solutions

Interactive 3D in the browser: product configurators, visualizations, CAD model viewers. WebGL, Three.js, no installs.

Software that solves real problems

Custom software, tailored to individual business and technical needs, is one of Nexebit’s core specialisms. The offer covers web and mobile applications, AI-powered systems, process automation, integrations and advanced 3D solutions.

The foundation is understanding the problem, choosing the right technology and a well-thought-out architecture. Performance, reliability and room for further development are key.

The result is solutions that streamline processes, solve specific problems and support business growth.

Selected projects

xeokit SDK

Open-source WebGL SDK for BIMDetails in portfolio

Problem: BIM and AEC apps need to display very large models in the browser, often placed at real-world GIS coordinates, with no plugins or installation.
Solution: A JavaScript and WebGL library supporting IFC 2x3 and 4.3, glTF, LAS/LAZ point clouds, BCF viewpoints and its own XKT format for fast loading of large models. Double-precision rendering and plugins for measurements, sections, the model tree and navigation.
Result: Used by Thinkproject, Fieldwire by Hilti, OpenProject and Deutsche Bahn, among others. The founder of Nexebit is a public contributor to xeokit-sdk and related repositories, including xeokit-bim-viewer and xeokit-convert.

Demo of a BIM viewer built with the xeokit SDK

In-house products

Automation

Own platform for delivering automations

Problem: Every company has its own processes, and a generic tool like n8n ends up as a tangle of nodes that is hard to test, version and maintain. Writing each automation from scratch means building the queue, schedules, secrets, logs and a client panel all over again.
Solution: A platform where an automation is plain TypeScript code and everything around it comes from the platform: organizations, configuration forms, schedules, webhooks, a job queue with retries, encrypted secrets, tables and a live step graph of every run. Workflows reach production from CI with a single command.
Result: Runs in production with in-house automations: finding leads and writing offers with Codex in EspoCRM, sending offers as letters, website monitoring and failure alerts. Each organization sees its runs, results and time saved in its own panel.

Step graph of a CRM offer generation run, with the offer written by Codex and saved to EspoCRM

Organization panel with its automations, runs this month, time saved and the next scheduled run

Website monitoring automation with its step graph and a table of checked sites with response times

Admin list of workflow deployments with the commit, SDK version and bundle size

CadSeed

Own product, not a client project

Problem: 3D printing solves many small problems, like an adapter between a vacuum cleaner and a pipe or a VESA mount, but designing such a part means learning CAD. Ready-made models from the web rarely fit to the millimetre.
Solution: A platform where the user picks interfaces from a catalogue or uploads their own model, and an AI agent helps choose and configure the operation. Geometry comes from a deterministic parametric engine (build123d), so dimensions are never guessed by a language model.
Result: A working MVP: from choosing two things to a print-ready STL, STEP or GLB file, with a 3D preview in the browser, a render queue and a chat with an agent about the model.

Adapter from a Dyson V15 vacuum pipe to a 27 mm Festool hose in the bridge editor

Bridge from a custom bracket model to a GoPro camera mount

Adapter from a 3/8-inch tripod thread to a VESA 75 plate

Raspberry Pi 4 case generated by the AI agent, with views of the attempt and the check result

Library of AI-generated and user-uploaded models

Have an idea for an app, an AI system or an automation?

Get in touch