ABOUT
URU Forge is an AI engineering company. We build autonomous agents, the LLM layer they run on and the wiring into your systems, and then we run them. A chatbot answers a question. A hive finishes a piece of work.
Why URU Forge exists
Most companies now have an AI pilot. Far fewer have anything in production, and the gap between those two states is not a model problem.
It is retrieval that was never tuned, evaluation that was never written, an integration nobody scoped, and an autonomy boundary nobody agreed until after an incident. Those are engineering disciplines. They are dull, they are learnable, and skipping them is why a demo that impressed a board in March is switched off by September.
We exist to do that part. Not the demo.
Four disciplines
AI development
We build the system itself: the agents, the models behind them and the products they live inside. This is where a job you can describe becomes a worker that owns it. → AI Development
LLM engineering
We engineer the layer underneath: retrieval, memory and context design, tool interfaces, model routing, guardrails and evaluation. It is what decides whether the system still behaves in month six. → LLM Engineering
AI integration
We wire it into what you already run, and we coordinate agents once there is more than one. Authentication, rate limits, retries and the failure modes nobody budgets for. → AI Integration
Autonomous workforce
We put it to work and keep it healthy: agents doing the work of skilled professionals, department by department, with someone watching the evals. → Autonomous Workforce
Five things we believe
Autonomy is a boundary, not a slider
An agent is not “more autonomous” than another. It has a written list of what it may do alone, and that list is enforced outside the model.
A system you cannot audit is a system you do not own
Every action logged, every decision reconstructable, and the log in your hands rather than ours.
The hard part is never the model
It is the data it can reach, the tools it may use and the exception it has never seen.
Automate the process, not the click
Rebuilding a bad workflow faster produces a fast bad workflow.
Some work should stay human
Judgment, negotiation and anything where a person has to be accountable to another person. We will say so at the gate rather than after the invoice.
How an engagement actually runs
We start with an audit that can tell you not to proceed, and we have written that recommendation more than once. Then a blueprint, then one agent measured on your real data, then a production build with the guardrails already in place.
You keep everything. The agents, the configuration, the eval suites, the runbooks and the audit history are yours whether or not we keep operating them, and the runbooks are written for a reader who is not us.
We would rather deliver one working agent than sell an architecture nobody needed yet.
Who owns URU Forge
URU Forge is the AI engineering division of URU Systems SMC-Private Limited.
That relationship is the reason this company started at the operations end of AI rather than the demo end. Running systems other people depend on teaches you what an autonomous agent needs before it is allowed near production.
Where we work from
We work as a distributed team and run engagements remotely by default, with the working hours agreed against your timezone rather than ours.
There is no office photograph on this page and no map, because neither would tell you anything about whether an agent will work.
Working here
We hire engineers who have operated something in production and remember what broke. If that is you, get in touch.