:%20tank%20med%20skvalpeskott%20—%20lagene%20som%20gjer%20agentar%20stabile%20--%3e%3crect%20width='400'%20height='400'%20rx='88'%20fill='%230f172a'/%3e%3c!--%20tankskal%20--%3e%3crect%20x='92'%20y='84'%20width='216'%20height='232'%20rx='36'%20fill='none'%20stroke='%23e2e8f0'%20stroke-width='14'/%3e%3c!--%20skvalpeskott%20(baffles)%20—%20tre%20lag,%20opne%20i%20midten%20--%3e%3cg%20stroke='%2338bdf8'%20stroke-width='14'%20stroke-linecap='round'%3e%3cline%20x1='118'%20y1='152'%20x2='186'%20y2='152'/%3e%3cline%20x1='214'%20y1='152'%20x2='282'%20y2='152'/%3e%3cline%20x1='118'%20y1='216'%20x2='170'%20y2='216'/%3e%3cline%20x1='230'%20y1='216'%20x2='282'%20y2='216'/%3e%3cline%20x1='118'%20y1='280'%20x2='186'%20y2='280'/%3e%3cline%20x1='214'%20y1='280'%20x2='282'%20y2='280'/%3e%3c/g%3e%3c!--%20væskelinje%20i%20ro%20--%3e%3cline%20x1='118'%20y1='120'%20x2='282'%20y2='120'%20stroke='%2364748b'%20stroke-width='10'%20stroke-linecap='round'/%3e%3c/svg%3e)
Open-source governance layer for autonomous AI agents
Use the intelligence. Keep your data.
Use AI agents with greater control over sensitive data and actions. Agents act only on what you permit, risky steps wait for a person, and every decision is traceable.
The shape of it
The open questions
Agents are capable. Control is the hard part.
What can an agent reach?
What may it do?
Where does the data go?
What happened, and why?
The trust layer
One place where every request is decided.
Security model
Five principles.
Local by design
Super Tanks runs inside your own environment, so sensitive data can be kept local when using a local model, rather than being sent to a conventional cloud-first service.
Default deny
An agent action that has not been explicitly permitted is refused.
Explicit permissions
Agents can only perform actions that have been explicitly permitted.
Full traceability
Activity is logged, so you can see afterwards what an agent did and on what basis.
Human approval for risky actions
Risky actions can be stopped and held at GO-Gate until a person approves them.
How it works
Ask, check, act — keep the record.
More detail
Super Tanks runs locally rather than as a conventional cloud-first service. Its security model is default-deny: an agent can only perform actions that have been explicitly permitted, and activity is logged so decisions can be traced afterwards. Sensitive data can be kept local when using a local model. Risky actions can be held at GO-Gate until a person approves them.
Under the hood
Built in the open.
Design partner programme
Introduce AI agents — with us alongside.
Likely a fit if you
- are planning to introduce AI agents
- have open questions on data security and governance
- need control over how agents reach sensitive information