The AI pilot never reaches daily work
The prototype impresses in the demo, but nobody dares to connect it to real data and real processes.
Outcome: A rollout with clear roles, boundaries and approvals that your team can stand behind.
AI agents · Infrastructure · Security
AI agents that take work off your plate – within clear boundaries.
I help leadership teams and engineering teams introduce AI agents, and the systems underneath them, so they become reliable colleagues: with defined roles, controlled access to data and results your team can verify and own.
- IndependentYou talk directly to the engineer
- Vienna & remoteOn-site in Vienna, remote across DACH, the EU and worldwide
- German & EnglishConsulting, delivery and documentation
When to call me
You want AI in production – without giving up control.
Most initiatives do not fail because of the model. They stall on open questions about data, accountability and operations. That is where I start.
Security and privacy slow everything down
IT, data protection or the works council rightly ask: what may the agent see, where does data go, who approves?
Outcome: An architecture that answers these questions before they become expensive.
Infrastructure that grew by hand
Servers and environments were built manually over years, and the knowledge sits in very few heads.
Outcome: Reproducible systems that can be reviewed, rebuilt and handed over to your team.
Your team lacks time or specialist skills
The topic is urgent, your engineers are fully booked and the experience is not in-house.
Outcome: Experienced support for a defined period that fits into your processes and tools.
Services
Four areas, one goal: systems you can trust.
I work where software meets infrastructure. That can be a new system, hardening an existing one, or simplifying operations that demand too much attention.
AI agents & LLM systems
I design and build AI agents and LLM integrations for concrete workflows – from research and analysis to internal tools. Every agent gets a role, a boundary and a point where a human approves.
- Agent workflows for real processes
- Isolated runtimes
- Controlled access to models and tools
Linux & NixOS infrastructure
I replace hand-maintained servers with declarative, reproducible systems. Changes become reviewable, rollbacks simple and handovers possible – on your own infrastructure, too.
- Reproducible servers and environments
- MicroVMs and containers
- Automation and rollbacks
Network & security
Threat models, segmentation, access and secrets management: I design systems for the failure you have to expect, not the one you hope to avoid.
- Threat models
- Segmentation and egress control
- Access and secrets management
Software & tooling
Command-line tools, services and APIs in Rust, Go, Python and TypeScript: software that operates systems and takes routine work off your team.
- Services and APIs
- Internal tools and CLIs
- Integrations with existing systems
The office in the background
Agents with roles. People with responsibility.
The scene behind this page shows how I think about AI agents in a company: as colleagues with a defined role in a shared office. Agents plan, research, build, test and operate. People set the goals and give the approvals.
AI agents by role
- Planning
- Research
- Engineering
- QA
- Security
- Operations
- Docs
People by responsibility
- Team lead
- Product
- Design
- Management
An illustration, not a live view.
Process
From the first call to the handover.
Clear steps, clear responsibilities. You always know what is being worked on, which decisions are pending and what comes next.
Intro call
You describe the goal, the starting point and the constraints. I tell you honestly whether and how I can help.
Clear scope
We write down goals, responsibilities, milestones and acceptance criteria.
Delivery in stages
I deliver in manageable steps and report regularly on progress, decisions and risks.
Handover
Documentation and onboarding, so your team can run and evolve the result on its own.
Ways to engage
Project
An agreed outcome with scope, milestones and a handover.
Hourly support
Targeted help with implementation, debugging, integration or architecture questions.
Team extension
I work inside your team, with your tools, and own agreed responsibilities.
Boundaries before convenience
Every agent gets only the access its task needs – enforced technically, not just agreed on.
Evidence over assertions
Done means verified. I show tests and evidence instead of an optimistic summary.
People approve
Decisions that matter stay with your team. Agents prepare, people decide.
Built for handover
You get systems your team understands and can continue – without a lasting dependency on me.
Open-source work
My work is open to inspection.
Part of my work is public. These projects show how I build systems: with clear boundaries and verifiable results. Each project's documentation states its maturity.
Open source · pre-1.0
Tentaflake
Runs AI agents in isolation on a NixOS server – declaratively configured, with credentials kept outside the agents.
Result: agents with their own boundaries instead of full server access.
Open source
Project Check
Consistent, explicit quality checks for software projects. “Ready” is never reported as “passed”.
Result: every check reports traceably as passed, blocked or not run.
Open source
Tabula
Project documentation that tells you when the code it describes has changed and a review is due.
Result: outdated documentation surfaces before anyone relies on it.
Open source
Ichnora
Local project memory for AI coding agents: decisions with provenance and evidence instead of loose notes.
Result: decisions stay findable across sessions, with evidence.
Open source
Visual QA
Evidence-based review of web interfaces with screenshots and accessibility checks; approvals stay human.
Result: every UI finding comes with a screenshot as evidence.
Open source
Commitell
Turns code changes into traceable, signed commits – fully offline if you prefer.
Result: a traceable change history without extra effort.
All public projects on GitHubTechnical deep dives on timwitter.dev
Who this is for
Answers for the people who decide.
CEOs & founders
You get a clear assessment of value, risk and effort – and a delivery your team can carry forward.
CTOs & tech leads
You get an engineer who connects architecture, infrastructure and security and fits into your workflow.
Recruiters & procurement
You get the facts at a glance: role, focus areas, location, languages and ways to engage.
About
Hi, I’m Tim.
I’m an independent engineer based in Vienna. I care about systems with clear boundaries: where an agent may act, what a server may trust and what a network should allow.
I translate between technology and business. Before I build anything, I want to understand the problem it should solve, who works with it and how much risk you can carry. Then I explain the trade-offs so that you can decide – not only your engineering team.
I like owning a well-defined piece of work and collaborating closely with your team. I am also honest about fit: for generic software that will be interchangeable within months, I am the wrong choice. I say so early.
Blog
Perspectives for decision-makers.
What AI agents need inside a company to work reliably – derived from my own open-source projects.
AI agents need budgets and boundaries, not just a prompt
An instruction is not an access control. How to separate agent budgets, permissions and acceptance evidence before a convincing demo becomes an operating service.
The hidden cost of “the AI says it worked”
A persuasive success message is not a test result. What evidence release decisions need, and why skipped checks must remain distinct from checks that passed.
Documentation that knows when to ask for review
Fresh files do not prove an explanation is correct. How linked sources, structural indexes and traceable decisions can direct attention to the next useful review.
FAQ
Short answers.
What does Tim Witter do?
Tim Witter is an independent engineer in Vienna. He designs and builds AI agents and LLM integrations, reproducible Linux and NixOS infrastructure, and the network and security layers around them – for companies, startups and engineering teams.
Do you work on-site in Vienna or remotely?
Both. I am based in Vienna and happy to work on-site there. Projects in Austria, Germany, Switzerland, the EU and worldwide I support remotely.
Which companies are a good fit?
Startups, SMEs and engineering teams in larger companies that run or introduce technically demanding systems – such as AI agents, infrastructure or security tooling. For generic software that is quickly interchangeable, I am a weaker fit.
How does a collaboration work?
After an intro call we write down the goal, scope and acceptance criteria. I then deliver in stages, report regularly and hand the result over to your team with documentation.
How do you stop AI agents from doing too much?
By limiting and technically enforcing access, credentials and network paths from the start. Agents run in isolation, see only the data their task needs, and a person approves decisions that matter. There is no absolute guarantee, so I name the remaining risks explicitly.
Can AI agents run on our own infrastructure?
Yes. Agents and their environment can run on your own Linux servers. Whether models run locally or through a provider is a trade-off between privacy, quality and cost that we decide together.
Which engagement models do you offer?
Scoped projects with an agreed outcome, hourly support and working inside your team as an extension.
What does a project cost?
That depends on scope, duration and engagement model. After the intro call you receive a concrete proposal with a clear scope.
Which languages do you work in?
German and English – in conversations, in delivery and in documentation.
Contact
Let’s talk about what you are planning.
Send me a short note: goal, timeline and preferred way to engage. You get an honest assessment of whether and how I can help.
- hello@timwitter.com
- Availability
- New projects on request – current capacity by email
Vienna · remote · German & English