FAQ
StableQuestions that come up in most evaluations of Pimsy.
General#
How is Pimsy different from a chat assistant with tools?+
A chat assistant responds to turns. Pimsy holds an objective across many steps, maintains state, revises its plan when reality disagrees with it, and produces verified artifacts. The difference shows up on work that takes hours rather than seconds.
Does Pimsy replace my CI, my reviewer or my judgment?+
No. It prepares work and proves it. Your pipeline stays the gate, your reviewer stays the authority, and accountability stays with you.
What happens when it does not know something?+
It reports the gap. The runtime is designed so that "I could not determine this" is an acceptable outcome, and a confident fabrication is a failure.
Safety and control#
Can a prompt injection make the agent do something dangerous?+
It can influence reasoning. It cannot grant capability. Policy is evaluated outside model context against signed configuration, so an instruction embedded in a web page or email cannot widen what the agent is allowed to do.
Can I stop a run mid-flight?+
Yes. Cancellation takes effect at the next step boundary, and completed work is preserved as artifacts. For repository work, applied patches remain on the working branch and are reversible via their derived reversal patch.
Where are my secrets?+
In a credential broker. They are injected at the transport layer, never placed in model context, traces, artifacts or logs.
Operations#
What does a task cost?+
It varies with steps, model routing and tool usage. Set a
usd budget — the run suspends and reports progress rather than overspending silently.How long can a task run?+
Long-running tasks checkpoint and resume across hours or days. The practical limit is your budget, not a session timeout.
Is my data used for training?+
No. Workspace data is not used to train shared models, and workspaces are isolated at the storage, index and embedding layer.
Last updated 2026-09-16