About
Built because coding agents changed what “safe secret handling” means.
ScopeHold started with a simple observation from working alongside coding agents every day: the moment an agent can read files and run commands, every .env file, shell export, and pasted key in a chat becomes part of its reachable surface. The habits that were merely untidy for humans became genuinely risky with agents in the loop.
Existing secret managers were built for apps and CI pipelines. They do that job well, but they hand the whole environment to whatever authenticates. We wanted something different: a credential layer where humans and agents are first-class actors, where access is scoped to the project an agent is actually working on, and where every retrieval leaves a trace.
ScopeHold is independent and built by people who run Claude, Codex, and Cursor on real projects daily. We use ScopeHold to hold ScopeHold’s own credentials.
What we optimise for
Scoped by default
An agent working on one project should never see credentials from another. Access is granted per project, per agent, per secret, not vault-wide.
Low friction or it will not be used
Security tools fail when they are annoying. The core job is simple: paste a secret once, choose a scope, use it from any approved place, stop thinking about it.
Auditable enough to trust
Reveals and retrievals are logged with who, what, and when. You should be able to show a client or a teammate exactly which agent touched which credential.
Nothing to take on faith
Secrets are encrypted with AES-256-GCM and per-workspace data keys, agent tokens are stored only as hashes, and the resolve CLI is open source so you can read exactly what runs on your machine.
Questions, feedback, or something we got wrong?
We read everything. Reach us through the contact options on the help page, or open an issue on the open-source CLI.