Industrial-grade AI engineering.
Agentas AS designs the architecture for AI systems, then builds and operates them — from industrial control systems to LLM-driven products, agentic automation, and custom applications, taken from brief to production.
Trusted by industry leaders
From AI systems to industrial control
Delivered for operators, EPCs, and system integrators on the Norwegian Continental Shelf — and for Norwegian industry onshore.
AI Architecture
Before you build with AI, decide how it fits: the use cases worth building, model choice, retrieval over your own data, evaluation, security boundaries and running costs. We design that architecture as your AI architect, then build and operate it if you want.
Read about AI architecture →
AI & LLM Engineering
Production LLM-driven systems built to run, not just to demo — agents, retrieval (RAG), tool-use, and continuous evaluation harnesses that measure output against real data. Designed for reliability, observability, and cost control.
AI-Augmented Automation
Agentic pipelines and integrations that replace recurring manual work — data ingestion, document processing, and scheduled jobs that feed your business systems. The automation discipline from industrial control, applied to software.
Custom AI Applications
Full-stack applications with AI at the core — Go, React, and TypeScript, self-hosted and secure by design. From internal tools to public products, built end-to-end and operated in production on our own infrastructure.
PLC Programming
Programming industrial controllers (PLCs) that automate and monitor equipment on Siemens platforms. We build custom control logic, reusable software libraries, and simulation systems for underwater and platform-based operations.
HMI / SCADA Development
Designing the screens and dashboards that operators use to monitor and control industrial processes. We build complete visual systems for real-time monitoring, alarm handling, and step-by-step operator guidance.
Subsea Control Systems
Complete software for controlling underwater oil & gas equipment — from the communication between surface platforms and seabed devices, to integrating valves, sensors, and hydraulic power systems. Proven delivery on active Norwegian Continental Shelf projects.
Industrial Communication
Making different industrial systems talk to each other. We set up secure, reliable data connections between physical equipment, controllers, and business systems using industry-standard communication protocols and cloud platforms.
Custom Software Development
Tailor-made software built to solve your specific problems. From automation scripts and data tools to complete web applications — using Python, Lua, and modern web technologies to fit industrial and business needs.
Commissioning & Startup
Hands-on support when systems go live — both on-site and offshore. We handle factory acceptance testing, on-site verification, equipment testing, and production startup to ensure a smooth handover from development to daily operations.
Private & On-Prem AI
Some data can't go to a public US cloud — a hard requirement across energy, healthcare, and the public sector, not a preference. We stand up open-weight models (Llama, Mistral, Qwen) that run entirely on your own hardware, behind your own firewall, with private retrieval (RAG) over your own documents. The same agentic engineering and operational discipline as our cloud work — your data never leaves the building.
Developer team starting with Claude Code? Setup and training →
Looking for something specific? See what we build, by outcome →
A proven track record of delivery
Projects delivered under Agentas contracts — control-system software for critical infrastructure and custom automation for industrial products. The founder's wider industrial record, from subsea control systems to hospital automation, is covered under About.
E1 Project — Oslo Water Supply
OneCo Technology
Building a backup water supply system for Norway's capital. Developing the control software and operator screens that ensure reliable operation of this critical city infrastructure.
Read the case study →
Turfpal — Sports Field Lighting
Turfpal
Custom software development for intelligent turf lighting systems. Control logic and automation for mobile sports field lighting rigs used across professional and municipal facilities.
Read the case study →
AI-first engineering — proven, not promised
At Agentas, AI is not a buzzword — it is how we work every day. We use Anthropic's Claude Code as our primary development tool, harnessing its full capabilities to write code faster, solve problems smarter, and deliver projects more efficiently. This website itself was built and deployed, and is maintained, entirely through Claude Code — living proof of what AI-powered engineering can deliver.
A selection of LLM-driven and AI-augmented systems Agentas builds and runs end-to-end on our own infrastructure — plus internal R&D projects that stress-test the AI-native workflow. Each entry focuses on the engineering — the back-end, the architecture, and the operational discipline — rather than the specific domain it serves.
Agentas Consult: AI-Graded Tenders & Offers
A multi-tenant SaaS platform, live since September 2026 and in pilot. It harvests every open Norwegian public tender each hour, grades each tender and job opening against the team's profiles with AI, checks mandatory bid requirements in code, and drafts priced offers exported to PDF or Word. Node.js, Express and one SQLite database per customer account, with 557 automated tests gating every deploy.
View details →Aurly: Ranked Trout Lakes for All of Norway
A consumer map app, live at aurly.no since September 2026, that covers more than 267,000 lakes in Norway and ranks the lakes in any area for brown trout by likely presence, size potential, remoteness and legal access. Built on data from 13 providers, most of them official open data, with rivers traced from the sea to model fish barriers, and every claim sourced. A Python and PostGIS data pipeline, a Next.js and MapLibre front end, Stripe and Vipps payments and 1,300+ automated tests.
View details →Agentas Suite: Ten Apps, One Column
The ten apps we run Agentas on, from mail, notes and hours to files, tenders and sales, built as separate apps that look and work like one. Every app carries the same icon column with keyboard shortcuts and counts of what is waiting, and a Claude side panel works beside whichever app is open. We build the same kind of suite for businesses.
View details →Mail: Two Gmail Accounts, One Inbox
A self-hosted web mail client on the Gmail API that puts a personal and a company account, aliases included, in one inbox and replies from the right address. Hostile mail is shown in a script-free sandbox with tracking pixels held back, and Claude can search, read and draft through MCP but can never send. Built and hardened in about eight days, with about 150 automated tests.
View details →Note: Offline-First Notes with Claude
A notes app for the browser and Android that saves as you type, works offline and merges edits from two devices without losing a word. It has the feel of Notepad++, Norwegian dictation on our own server and an MCP server that lets Claude read and write notes. TypeScript, CodeMirror 6 and Yjs, with about 175 automated tests.
View details →Agentas Timesheet: Hours in One Tap
A timesheet app for the phone and the PC that replaced a spreadsheet. A ready-made day is confirmed in one tap, Norwegian holidays, flex and overtime are worked out in tested code, and Claude can draft the day from the work that was done for a person to confirm. Works offline, with passkey sign-in.
View details →CloudDrive: Self-Hosted File Management
Self-hosted file management platform (Go + React) built around a deliberate security model from day one: access control, audit logging, encrypted backups, and regularly rehearsed recovery. Designed so that "self-hosted" is safer than the cloud alternative, not an excuse to be sloppier.
View details →Agentas WebOps: Internal Ops Panel
The internal panel that runs our website arm, Agentas Web, end-to-end: client and site registry, an accounting-driven billing watch that auto-suspends non-paying sites via an nginx maintenance map, and a full leads engine with public-registry discovery, enrichment and scoring, drip outreach with hard daily caps, and a deal pipeline through to invoicing. Express + built-in node:sqlite, zero native dependencies, one process. The unglamorous machinery that makes the studio run itself.
View details →Vehicle Telemetry Automation
An automation pipeline that pulls live vehicle telemetry into a structured database, then runs scheduled jobs that feed mileage and timesheet entries to downstream business systems. Replaces a recurring monthly manual task entirely — small in scope, real in time saved.
View details →Smart Home Automation
Self-hosted smart-home platform integrating Z-Wave, Zigbee, Bluetooth and Wi-Fi devices, scheduled automations, multi-room control, and energy monitoring. All configuration version-controlled and deployed through an automated pipeline. The same engineering discipline is applied at home as in client systems.
View details →Automated Trading Platform
An automated trading platform spanning crypto and equities: a strategy engine with backtesting, execution simulation and a web cockpit, all sharing one code path proven identical across backtest and the running bots to within a rounding error. 320+ automated tests, a deterministic decision core (no LLM in the trade loop, by design), and every result reported against random-baseline nulls. Runs as several containers from a single image.
View details →AI-Assisted Market Analysis
A research platform that ingests official company disclosures, classifies them with an LLM, and puts every candidate signal through rigorous statistical validation — null-hypothesis baselines, permutation gating and a realistic cost model — before any number is trusted. Built to test market hypotheses honestly: measure first, believe later. The engineering is the point — disciplined data pipelines and statistical rigour over hype.
View details →Watcher: LLM-Graded Watchlists
A self-hosted watcher that follows a set of sources, grades every new item against what you are looking for with an AI model, and ranks the results on one dashboard. Every model call runs in a sandbox with every tool denied, and all external text is fenced as data, not instructions. Prompt injection treated as an operational reality, not a footnote.
View details →Table-Scout: AI Restaurant Finder
An LLM-assisted local restaurant finder — "what's open now, book a table for four at seven." It ingests open geodata and the official business registry, matches and geocodes venues, detects booking providers, and answers in the asker's own language. Wired into a smart home as a voice agent, and built with careful data-licensing hygiene: open-data columns kept structurally separate from own-collected data.
View details →AssistKey: Push-to-Talk for Home Assistant
A Windows system-tray push-to-talk client for Home Assistant Assist — hold a hotkey, speak, get an answer. Deep native integration: DPAPI-encrypted token at rest, per-user autostart via the registry, dark-titlebar and monitor-aware window handling, single-instance control, and an optional wake word. Open source, packaged as a single-file executable with CI on Windows.
View details →Cloud Infrastructure
A multi-server environment evolved over two years into our own small cloud infrastructure, with 30+ services across web apps, automation, monitoring, and home systems. Standardized deployment pipeline, layered defense-in-depth security, three-tier backups with rehearsed recovery, and a living architecture reference maintained like a runbook.
View details →Multi-Tenant Analytics Platform
A free analytics platform for a player community: uploaded logs are parsed on the server in worker threads into fight-by-fight breakdowns and rankings, stored in a normalized relational database. One codebase serves each tenant from its own container behind a shared reverse proxy. Built and operated single-handedly.
View details →Go-based Computational Simulator
A computational simulator written in Go with a browser-based UI and a continuous evaluation harness that compares simulator output against real observed data — surfacing discrepancies explicitly rather than papering over them. Built on a multi-class architecture with a shared widget system across simulators.
View details →NoBS: Workout Log
A no-BS gym app: only the stuff you use at the rack: sets, reps, kg, a rest timer, heart rate. Local-first, no account, no subscription. Built as a React + TypeScript PWA, packaged for Android with Capacitor and published on Google Play, with its own landing page and a direct APK download.
View details →Baby Suite: Three Local-First PWAs
Three installable web apps built around our baby's arrival — a contraction timer, a feed/sleep/diaper tracker, and a postpartum pelvic-floor trainer. Each app runs a sync conflict model deliberately chosen for the data it handles, with hand-rolled dependency-free Web Push (VAPID + payload encryption in raw WebCrypto), server-side alert rules on a cron, and free-tier quotas treated as a hard design constraint. In daily family production — where an outage means a missed 3 a.m. feed alarm.
View details →Dreadmark: Isometric Action-RPG
A top-down isometric action-RPG built from scratch in Unity 6. An AI coding agent writes and tests the code, a human designer sets the direction and judges every build, and bots that play the game check each release. A free demo is out on Steam.
View details →Frostkeeper: 3D Survival Game
A top-down survival game in a frozen valley, in early development in Unity 6. Its own weather, fire and night light are built and judged one look-test pass at a time, with the same AI-augmented workflow used across the rest of the work.
View details →Click any card for details and screenshots. Several projects are live and linked; the rest are kept private — code walkthroughs and demos are available for serious inquiries, just get in touch.
AI-native software engineering, grounded in industrial rigor
Agentas AS is a Norwegian software engineering consultancy led by Martin Davidsen. We design, build, and operate software the AI-native way — LLM-driven systems, agentic automation, and full-stack applications, shipped end-to-end and run in production. AI is not an add-on to how we work; it is the method.
That practice is built on a strong track record of delivering software for industrial and critical systems — subsea production control, offshore platforms, and operator tools across the energy sector. Based in Rogaland, with a background in maritime electrical automation and certified electrician credentials, we understand systems from the field hardware up to the software that runs them — and we bring that same discipline to every AI system we ship.
AI-Native Delivery
Every project is designed, built, and shipped with AI in the loop — from prototype to production and ongoing operation on our own infrastructure.
Certified & Trusted
Certified electrician with Siemens expertise and offshore commissioning credentials.
Industry Veterans
Track record with Siemens, Siemens Energy, and major Norwegian infrastructure projects.
Who you'll work with
Agentas is led and operated by Martin Davidsen — a Norwegian software and automation engineer with eight-plus years of experience building software for complex industrial environments. Originally trained in maritime electrical automation, with certified electrician credentials and offshore commissioning experience, and a track record across Siemens Energy and Aker BP subsea developments, OneCo automation projects, and major Norwegian infrastructure work.
Alongside industrial delivery engagements, Martin builds and operates Agentas' own LLM-driven systems end-to-end on company infrastructure. Real users, real traffic, real consequences when something goes wrong. That work is showcased in the AI & Innovation section above.
The combination — industrial discipline paired with hands-on AI engineering — is what Agentas brings to every project.
Let's discuss your next project
Whether you need expert automation engineering, custom software, or guidance on integrating AI into your workflows — we're ready to help.