[ Security ]
Your team’s knowledge stays under your control.
An overview of how we protect your knowledge and the agents that work with it. If your review needs more detail, ask us and we’ll answer directly.
[ Our posture ]
Built so you stay in charge of what agents can reach.
Each team’s knowledge is kept apart
Everything is scoped to a workspace owned by one organization. Nothing is shared or searchable across them.
People approve every agent
No agent reaches your knowledge until someone on your team allows it. Every connection is listed, and you can revoke any of them at once.
Every change can be undone
An edit keeps the version before it, and deleted work goes to a trash you can restore from.
Automation runs where you allow it
Agents started by Markbase run on your own machines, in folders you have approved, with only the access you give them.
Encrypted in transit and at rest
Connections use TLS, and the database, files and backups are encrypted where they are stored.
Your content isn’t used to train models
We don’t train models on your knowledge, and we don’t sell or share your personal information.
Inbound addresses expire on their own
The email addresses and webhook URLs an agent makes are short-lived, and only the person who asked for one can see it, not even an admin.
What arrives is treated as data
Agents are told to read forwarded emails and webhook requests as data, never as instructions. A webhook URL can also require a secret.
Webhook content doesn’t linger
A request’s content is deleted an hour after its URL expires. Markbase never logs the URL, its secret, headers or bodies, and never forwards them anywhere.
Jobs only go to machines that can run them
Each automation names a profile: the agent, model, effort and time limit it runs with, and which machines may take it. Machines check what they can actually run, and a job is only handed to one that matches.
An optional sandbox for each job
A profile can run its agent inside Anthropic’s sandbox runtime, which limits the files it can read and write and the domains it can reach. Markbase sets that policy, not the agent.
Automatic choices send a short summary
Only a profile setting left on automatic is chosen by a third-party classifier, listed on our subprocessors page. It receives the start of the job’s prompt, the alert’s main fields and your lane names, never secrets, tokens or attachments. Self-hosted installs use their own key.
Who can see a job follows the workspace
A job’s page shows its prompt, the alert’s content and the agent’s output, so a job that belongs to a workspace is visible only to people with the right access to it. A job with no workspace is visible to members who can reach at least one workspace.
Live activity is off until you turn it on
A profile can show what its agent is doing on the job’s page. Known secret formats are removed on your machine before anything is sent, and the activity is stored encrypted and deleted after 14 days.
Email jobs check who sent the message
An address limited to your team or a list you set only accepts mail that passes DMARC. If you let anyone send, each job waits for a person’s approval by default, and mail that fails DMARC is always refused.
[ Where it runs ]
On established cloud providers, or inside your own.
The service runs in Germany and keeps files in the United States. Every provider that handles your data is on our subprocessor list, with what it handles and where.
When your requirements rule out a shared service, run Markbase as your own deployment.
[ Higher security requirements ]
Keep Markbase inside your own boundary.
Some teams can’t send their knowledge to a shared service, however it is protected. For them Markbase runs as its own deployment, with every Enterprise feature, so your data stays in the environment your policies already cover.
You run it
Self-hosted
Markbase inside your environment, on premises or in your cloud, operated by your team. Your knowledge, its history and your agents’ jobs never leave the perimeter you already defend.
- Knowledge, skills and automation are delivered from your install, not through our hosted service
- Reviewed, configured and monitored under your own security policies
- We only see error reports with user identity and request contents removed, which you can turn off, and what you choose to share with us for support
We run it for you
Dedicated
A single-tenant deployment in your own cloud account, operated by us. Your knowledge stays in your account and the region you choose, without your platform team taking on another service.
- No other customer shares the deployment
- Your account, your region, and the provider you already approve
- We never delete data from your infrastructure without your written instruction, and our access ends at the agreed handoff
[ Commitments ]
What we promise in writing.
- GDPR and CCPA
- Our DPA covers both, with Standard Contractual Clauses for transfers from the EEA, UK and Switzerland.
- Breach notification
- If a breach affects your data, we tell you without undue delay and within 48 hours.
- Subprocessor changes
- We give 7 days’ notice before adding one, and you can object.
- Access to production
- Our own access needs multi-factor sign-in, individual accounts and only the privileges each job requires.
[ Found a vulnerability? ]
Tell us privately first.
Write to hello@altacoda.io. We reply within two business days, keep you updated while we fix it, and won’t take legal action against research done in good faith.
Have questions for your security review?
Send us your questionnaire or ask us anything, including the technical detail this page leaves out. You’ll hear back from the people who build Markbase.