Notes
Notes
GembaOS is a zero-trust governance runtime for enterprise AI agents: policy-decided access, single-use clearance tokens bound to exact parameters, and accountable human approvals, running on-premises or in a private VPC. The notes below explain decisions and records for agents and people, including how to start with people how a browser demonstration can become a reviewed rule draft, and how the people who answer for a rule find out whether it works before they change it. Each note states the evidence and limits relevant to its subject; the glossary defines the recurring terms.
- Does the rule work? Finding out from the cases, and changing it under a nameHow GembaOS finds out whether a rule works - sampled answers, outcomes after a case closes, drift - and changes it only through a release a second person approves.
- Show the work once, and get rules someone answers forAn employee does the task once. The assistant writes down what it understood and asks, the rule's owner confirms, and a second person approves the release.
- Start with people, add the agent one step at a timeA business can enter GembaOS with no agent in it: people read, propose, sign and carry it out by hand. The same playbook hands one step to an agent later.
- Describe how your company works, and get rules that actually runWrite the rules yourself, or describe in plain words how the company works. GembaOS compiles one release draft, checks it, and a second person approves it.
- The agents ran all week. What does the manager open on Monday?Six console pages a manager reads after agents have run: the receipt, the overview, the agent registry, a checked rule release, the risk register, the connections.
- An agent works on a task for three days. At six in the evening, who is handling it?Anyone on the team can watch a running agent in GembaOS. One named handler can steer it or hand it to a colleague, and none of that moves the signature.
- Who closes the loop in an AI operating system for companies: the system, or a person whose name is on the record?YC asks for a company legible to AI by default. GembaOS explains why enterprise data classifications, privacy, and audits require closing the loop on a name.
- A vendor shipped a breaking change with no warning. What is on your desk in the morning?When an external API changes, code rarely crashes. GembaOS probes sandboxes nightly, maps business impact, and requires dual sign-off before engineers merge fixes.
- A day in the life: eight moments where an agent was not allowed to decide aloneOne working day at a company whose AI agents do real work: eight moments where the agent proposed and did not decide, each a step of a demo you can try to break.
- How do you prove what your AI agent is allowed to do — and what changed last month?The approval route an AI agent follows is printed from the configuration that runs, and two versions compared with each difference labelled looser or stricter.
- Your chatbot's answer is your company's promiseA tribunal held an airline to what its website chatbot said. What that means for AI agents that answer customers, and where a person signs before a promise is made.
- Why was this lot shipped? Structuring evidence for manufacturing concessionsConcession decisions happen daily in manufacturing, but evidence stays scattered across emails and travelers. Here is how GembaOS binds MRB approvals to releases.
- How should an AI agent ask a customer for missing documents?Chasing documents is where claims and vendor reviews lose weeks. An agent can help if it asks by kind, on a due date, and the case waits on the work item.
- Why Enterprises Hesitate to Let AI Agents Touch Real MoneyEnterprises do not fear AI capability; they fear unallocated liability and rogue execution. How intern models and single-use clearance tokens solve it.
- AI agent idempotency: how to stop one approved action becoming two writesAgents retry. Idempotency decides whether a retry is the same write or a second refund. Where the key comes from, and why the agent must not be the one choosing it.
- Stale approvals: valid when requested, but is an approval clicked hours later still safe?An action is valid when requested, but reality drifts during meeting delays. GembaOS re-reads live state right before writing, stopping stale approvals.
- What is a clearance token? Single-use execution permits for AI agentsA clearance token is a single-use, time-limited permit GembaOS issues for one approved AI agent action, bound to its exact arguments. Definition, lifecycle, FAQ.
Last updated: · GembaOS v0.2