Start with the boundary.
Security is an architectural requirement, not post-incident decoration.
THE FOUNDER / ENGINEERING MANIFESTOThe answer to a broken system
The answer to a broken system
is a working system.
Clawsolator exists because the founder needed secure, isolated agents for real work and found the available setup insufficient for his requirements. It is a product shaped by direct use—not a committee exercise.
THE FOUNDER’S POSITION UNFILTERED / EDITORIAL
“I don't give a fuck about consensus unsupported by experiments.”
Debate the theory all day. Somebody still has to build the thing, install it, break it, repair it, measure it and put it back into service.
That's how this began. The founder needed autonomous agents for engineering work. He needed isolation, protected access, sane approval boundaries and resource governance. The tools available were useful, but they did not present the complete experience he required. So the missing pieces became a product.
Our working rule is simple: reuse established technology when it earns its place. Replace it when it doesn't. Invent only when a real gap remains.
And if a model burns cycles, RAM and tokens to solve a problem a smaller mechanism can handle correctly, measure the waste. Don't romanticize it.
One day the underlying harness may be completely rewritten. It may be OpenClaw today, Hermes tomorrow, a XERXES-native system after that. The control architecture must survive that replacement.
XERXES SI FOUNDER'S ENGINEERING PHILOSOPHY
REPRODUCIBLE BY DESIGN
We publish what we use.
The project is intended to document its design choices, failure modes and demonstrations so other engineers can learn, reproduce results and challenge the evidence.
A green light is not proof.
Distinguish what was configured from what a container, network or daemon actually enforced.
From agents to synthetic intelligence.
SIDE is intended to host an iterative, measurable development cycle beyond today’s LLM harnesses.
THE NEXT STEP
Bring evidence. Not applause.
The strongest argument for new technology is a reproducible result.
FOUNDER / PUBLISH WHAT WE USEOPINION ENGINEERING PROOF
Build the answer. Test the answer. Publish the evidence.
Clawsolator grew from a practical constraint: the founder needed tool-enabled agents without giving the model unrestricted trust in a personal and business environment. Instead of treating the agent’s own assurances as protection, the work moved toward independent daemon-governed resource decisions, isolated runtimes and observable approvals.
“Stop paying for computation that leads nowhere.”
The founder argues that unnecessary tokens, memory use and tool loops should be measured and eliminated, not disguised with more infrastructure. Synthetic-intelligence research is a long-term direction, not an assertion that a general AGI has already been demonstrated.
One bounded request. One verified outcome.
A 32 MiB increase to an agent container was authorized by daemon policy, verified through Podman readback and journaled without an operator prompt or container restart. That is a useful specific demonstration—not a universal security guarantee.