Security
Built so one company can never see another's data.
PlurelOS holds your boards, meetings, deals and diligence files. We designed it so the walls between workspaces live in the database, not in good intentions.
Four layers between a request and your data.
Pick a layer to see what it does. Every way in, from the app to an AI client, passes through the same stack.
Every way in
AppAPISearchCompany BrainMCPSlackTenant wall
Every row in the database carries its workspace, and Postgres row-level security refuses any query that reaches across. It is built so a mistake in application code still cannot read another company's rows.
- Row-level security on workspace data
- Per-workspace encryption keys
- Dedicated database and region option for larger teams
Controls built into every workspace.
Most of these are on for every workspace. The ones marked Admin controls are for larger teams that need tighter control.
Row-level isolation
Postgres row-level security scopes every query to your workspace.
One permission check
App, API, search, AI and MCP share a single check.
Per-workspace keys
Each workspace has its own encryption keys.
Secrets stay out of AI
Tokens, keys and passwords never enter model context.
Throwaway sandboxes
Code runs in an isolated sandbox that is discarded after the task.
Append-only audit log
A record that can be added to, never edited. Kept for up to 1 year.
SSO and SCIM
Single sign-on and automatic provisioning via WorkOS.
Roles, approvals, conditions
Custom roles, approvals, IP, device and hours rules, and access reviews.
Export anytime
Take everything with you in open formats, whenever you want.
AI and your data
Helpful, and on a short leash.
Company Brain, live notes and AI briefs follow the same rules as everything else, plus a few of their own.
- What it can read
- Only what the person asking can already see. In a shared Slack channel, only what every member of that channel can see.
- What it can do alone
- Draft, summarize and suggest. It never sends, shares or deletes anything without asking you first.
- What it never sees
- Secrets. API tokens, keys and passwords are kept out of model context entirely.
- Where code runs
- In throwaway sandboxes that are created for one task and then discarded.
- Who processes it
- Anthropic's Claude models and the other AI models we use. We call every one of them only through OpenRouter, under zero-retention terms, so prompts and answers are not kept after the request. Your content is never used to train models.
Compliance, honestly.
We are not SOC 2 certified today. Here is where things stand. When something changes, this page changes with it.
The architecture on this page
Row-level isolation, one permission check, per-workspace keys and an append-only audit log.
SOC 2 Type 1
We will post the target date here once the audit is booked.
Independent penetration test
Timing and a summary of the results will be posted here.
Subprocessors.
The companies that help us run PlurelOS, and why. Last updated October 22, 2026.
| Subprocessor | Purpose | Data involved |
|---|---|---|
| Cloudflare | Hosting and file storage | App traffic and uploaded files |
| PlanetScale | Postgres database | Workspace data |
| WorkOS | Sign-in, SSO and SCIM | Names, emails and directory data |
| AssemblyAI | Meeting transcription | Meeting audio and transcripts |
| OpenRouter | Routes every AI request, zero retention | Prompts and context for each request |
| Anthropic | AI features, reached only through OpenRouter, zero retention | Prompts and context for each request |
| Resend | Product, sign-in and post email; mailing lists | Names, email addresses and post content |
| PostHog | Product analytics | Usage events, no workspace content |
Responsible disclosure
Found something? Tell us first.
Email contact@plurelinc.com with steps to reproduce. We will confirm we got it and keep you posted while we fix it.
Please
- Test only against workspaces you own
- Never access or keep other people's data
- Give us time to fix before you share details