Zero-Access Security
Vecta's hosted control plane coordinates jobs; the agent machine connects to your source and destination. The control plane needs job and run metadata, but does not receive backup data or the credentials the agent uses to access it.
Information stored by the control plane
Vecta stores the information needed to display and operate jobs:
- Account email, agent names and IDs, installed versions, and check-in times.
- Job names, schedules, destination URLs, and filesystem source paths.
- For PostgreSQL jobs: host, port, database, username, and SSL mode.
- Selected hook IDs and their configured parameters.
- Safe agent capability metadata: available features and hook IDs, descriptions, phases, parameter schemas, and whether a hook is allowed to run as root.
- Run status, stage details, duration, progress, processed/transferred counts, snapshot IDs, and the PostgreSQL archive filename.
Destination URLs are visible because the dashboard needs to show where a job writes. Treat database connection settings, hook IDs, and hook parameters as metadata visible to the control plane. Do not put passwords, tokens, or other secrets in hook parameters.
Information kept on the agent
The control plane does not receive:
- Restic repository passwords or destination access keys.
- SSH private keys.
- PostgreSQL passwords or database credential files.
- Filesystem backup contents or PostgreSQL dump contents.
- Hook executable paths, argument templates, or script contents.
The agent reads the source and sends backup data directly to the configured destination. Its local credential files remain on that machine. Protect an agent like any server that can read your source data and access your backups.
Hooks and trust
Hook definitions live in an administrator-controlled local catalog. A job can select an exposed hook ID and provide values allowed by that hook's parameter schema; the backend does not receive the executable itself. However, a hook is still an operation that dashboard users can ask the agent to perform. Only register hooks with effects and inputs that are appropriate to expose. Review the root privilege and local account assigned to each hook.
One-time secrets
The repository password and agent API key are shown once. The registration token is single-use and expires after 24 hours.
What this boundary means
In normal operation, the control plane stores job metadata but not backup contents or the credentials on the agent. However, the agent pulls job settings from the control plane. Someone who fully compromises that control plane may be able to change settings and ask an agent to back up a different source, use a different destination, or invoke a hook exposed to jobs. Local hook validation prevents arbitrary executable paths, but it does not prove that every remote job setting is trustworthy. Treat the control plane and agent as part of the trusted system; limit hook effects and secure both.
A compromised agent is different: it can access that agent's sources and destination credentials. Capability reports help the dashboard check compatibility; the agent still validates each job and hook locally.
What's next
Read The pull-based agent model to understand the network boundary, then Database Backups and Hooks for local setup details.