Plain answers.
The questions teams ask before a pilot. If yours is not here, ask us.
Can we run Obsilient ourselves?
The engine is delivered as a service, so the only thing you install is a small, read-only connector. For teams that cannot send anything out of their environment there is an on-premises option: a signed container under a time-limited licence. It is a service, not a toolkit to copy.
What happens when we stop?
The connector comes out with one command, what it sent is deleted within the window agreed at the start, with written confirmation, and you keep every report and finding. The licence lapses and the gate switches off without blocking your deploys. More on the end of an engagement.
Which models and frameworks does it work with?
Any model: hosted, open or your own, because Obsilient reads how your system is built and run, not what the model says. Today it reads agent systems written in Python, whether built on a framework or on plain code. If yours is built differently, we will tell you in the first conversation whether it fits.
Does it change our system?
No. The connector is read-only and never runs your code or your agents. The one thing you add is the gate on your change path, and you decide whether a fail blocks a change or only warns. Every fix stays your decision.
What does the connector send?
The structure of your system, which components exist and how they connect, and counts of how much work passed through each. Never source code, prompts, model outputs, your data or your credentials. You can have it write what it would send to a file first. Details on the security page.
Do you need access to our production data or model outputs?
No. The connector reads code, configuration and run logs where they already are, and sends only structural facts and counts.
How soon do we see something?
Within the first week of the pilot: a map of the system and the first ranked findings. The walkthrough follows, and the gate is in place by the end of the pilot.
What does a pilot need from us?
One system to look at, a person who knows it, permission to install the read-only connector, and time for the walkthrough. Your security team’s questions are answered in scoping, before anything is installed.
How is it priced?
A fixed-fee pilot of about a month on one system, which you can start online and which is credited against the first year if you continue. After the pilot, an annual subscription per system in one of three tiers, Map, Gate or Continuous, invoiced. Talk to us and we will scope it with you.
Is this another tracing or evaluation tool?
No. Tracing and evaluation tools show what happened on the runs you instrumented. Obsilient finds the failures those runs never show: the parts that were built and never connected, work that quietly stops between agents, and checks that pass when they should not.
Run agents in production?
We are working with a small number of early teams. Tell us about your system and we will tell you plainly whether Obsilient fits.