Client build · Several industries · Private AI agents · Sole developer and security architect
OpenClaw deployments
Private AI agents running for five businesses
OpenClaw is an open-source AI agent. Logo from the OpenClaw project.
5
businesses running a private agent
1 each
separate instance, keys, and data
Before
OpenClaw is a capable open-source AI agent. Out of the box it can reach more of a company's systems than most owners are comfortable with, and CrowdStrike and Nature have both written about the risks.
After
Each business runs its own copy of the agent, set up for its work, with clear limits on what it can touch and a log of what it did.
What we built
A separate instance for each client, on their own machine or a private cloud server
Tools, workflows, and prompts tailored to each business
Limits on which actions the agent can take, with block and allow lists
Activity logs so the owner can see what the agent did
Each deployment is a hardened fork with data isolation, per-client key management, and prompt-injection defenses.
Restricted actions and limited access shrink what a mistake or attack could reach. Audit logs and monitoring cover the rest.
These controls reduce exposure. No setup makes an AI agent immune to attack.
