Security · secrets, access, actions
How Spiron keeps your secrets and data safe.
How your agents use passwords and API keys without seeing them, how access is checked, how your workspace stays separate, and how every important action is confirmed.
- Website
- https://spiron.ai
- API
- https://api.spiron.ai
- Security contact
- security@spiron.ai
On this page6 sections
Secrets
Your agent uses a secret's name, never its value.
When you give Spiron an API key or token, the AI only ever sees a name like GITHUB_TOKEN. The real value is added at the last moment, only to the step that needs it.
Secret values never appear in what the AI reads or writes: not in prompts, conversation history, tool output, logs, or search. If a tool prints a known secret by accident, it is removed before the AI sees the output.
exec(command="gh api /user", credentials=["GITHUB_TOKEN"]) # The AI names GITHUB_TOKEN. The value is provided to this # one command only, and is never printed back.
- Encrypted at rest
- Secret values are stored encrypted.
- Missing secret, no run
- If a secret is missing, the step fails clearly instead of running without it.
- Lockable
- You can lock a secret so an agent cannot change or delete it.
- Logged without the value
- Every create, update, delete, and use is logged with who, what, and when — never the value.
- Rotate or revoke anytime
- Secrets can be rotated, revoked, or deleted from your dashboard.
Access
Permission is checked every time.
Being signed in is not the same as being allowed. Every request and every use of a secret is checked against your workspace's rules.
Identify
Every request to Spiron is tied to a signed-in account.
Check
Spiron checks whether that account, or that agent, is allowed to do this specific thing.
Allow or deny
If the answer is unclear, the request is denied rather than guessed.
Record
The use is logged. Permissions you revoke stop the very next attempt.
Sharing is always explicit
Letting another agent or teammate use a secret takes an explicit permission from its owner. Copying a workspace or project never copies access.
Isolation
Your organization is walled off.
Every agent, conversation, memory entry, and action belongs to one organization — and nothing crosses that line by default.
- Separate by design
- Workspace boundaries are built into the data itself, not applied as a filter afterwards.
- Separate inside your team too
- One agent cannot read another’s secrets or conversations without permission.
- No exceptions
- There is no built-in account or background service that skips these checks.
- Retries stay in bounds
- A retried or resumed job keeps the same organization and permissions.
Actions and evidence
Every important action is confirmed.
Sending a request is not the same as getting a result. Spiron checks that each important action actually happened.
Allowed
The action is checked against your permissions and budget before it runs.
Attempted
The action runs. A retry never sends the same email or payment twice.
Confirmed
Spiron checks the result with the system that was changed — the payment provider, the chat app, the database.
Recorded
The confirmed result is saved as a record. If confirmation never arrives, the action is shown as failed.
Rules
What Spiron will not do, even when it would be convenient.
- No self-approval
- An agent cannot approve its own high-impact changes.
- No going over budget
- An action that would exceed your budget or leave your workspace is not attempted.
- No asking twice
- A denied permission is not retried in the hope of a different answer.
What we do not claim
No badges we have not earned.
This page describes how Spiron works. It is not a certification, an audit report, a penetration-test result, or an uptime guarantee, and we do not claim any of those here. If something is not stated on this page, treat it as not claimed.
Reporting a problem
Responsible disclosure.
Vulnerabilities, suspected leaks, and security questions go to one monitored address.
Security
security@spiron.ai
Report suspected vulnerabilities to security@spiron.ai. Please do not post vulnerabilities, credentials, or customer data in public.
Tell us what is affected, when you noticed it, what you expected versus what happened, and enough detail for us to reproduce it. If you think a credential has leaked, rotate it first, then tell us.
- What to expect
- We aim to confirm receipt within 3 business days and to tell you how severe it is within 10.
- What not to send
- Live credentials, customer data, or a working exploit in a public channel.
Scope, testing rules, and our safe harbour for good-faith research are in the vulnerability disclosure policy. The same contact is listed on the trust page and in /.well-known/security.txt.
| What | Address |
|---|---|
| Website and dashboard | https://spiron.ai |
| API | https://api.spiron.ai |
Next step
Read the trust page.
It covers spend caps, prices, and how your data is stored and deleted.