Security at Kaiva Bridge
How Bridge protects your credentials, your data and every tool call your agents make.
AES-256
Stored credentials are encrypted at rest.
Per workspace
Isolated at the database layer.
Hash-chained
Every call is written to a verifiable log.
Tool digests
A changed tool changes the digest.
Read-only by default. Writes when you decide.
Every tool starts switched off, and each one is marked as a read or a write. Database, spreadsheet and document servers can only read: their queries run in read-only transactions, so a tool call cannot change your data.
When you want your AI to do more
APIs can do more than read, so Bridge lets you add writes deliberately, one tool at a time, with guardrails around each one.
- Turn on write tools one at a time, after you have reviewed them
- Require a person to approve a write before it runs
- Try writes in test mode first: writes are simulated, reads are real
- Every tool call, read or write, is in the audit log with its arguments and result
Where each control sits
Tool-set attestation
Every server carries a digest of the exact tool set it serves: names, descriptions and input schemas. If any of them changes, the digest changes.
Change detection, not just a tool count
A tool that keeps its name and has its description or schema rewritten is reported as a contract change, separately from tools being added or removed. That is the case a connected client cannot notice for itself.
Tamper-evident history
Each change is recorded with a timestamp, who caused it and a hash of the one before it. Editing or deleting an entry breaks the chain, and the check is a request you can make yourself.
Rather than imply certifications we don't hold, here's the honest position, so your security team can plan around it.
- GDPRAligned
Aligned. Data Processing Agreement available on request.
- SOC 2 Type IIIn progress
In progress, not yet certified.
- ISO 27001Planned
Planned, not yet audited.
DPA and sub-processor list available on request, email [email protected].
Found a vulnerability?
We investigate every report. Email [email protected] with details and steps to reproduce, and give us reasonable time to remediate before any public disclosure. Our policy is published at /.well-known/security.txt.