Security
Your contracts stay in Europe.
Every system that touches your contract text runs inside the EEA, operated by European companies. This page lists them by name. The binding version is the Data Processing Addendum — if anything here and the DPA disagree, the DPA wins.
Where your data lives
The platform runs on Clever Cloud in France. Contract files are replicated to OVHcloud in Germany for disaster recovery, stored there as sealed ciphertext. Both are European companies operating European infrastructure, which is what removes US CLOUD Act exposure — not a contractual promise about it.
Every sub-processor, named
These are the essential sub-processors. Essential means the platform does not work without them, so there is no opt-out. All processing happens within the EEA.
| What it does | Company | Location |
|---|---|---|
| Hosting, database, object storage | Clever Cloud | France |
| Encrypted replication and backup | OVHcloud | Germany |
| Transactional email and notifications | Sweego | France / Netherlands |
| Logging, tracing, error monitoring | Better Stack | Czech Republic |
| Default AI model provider | Mistral AI | France |
These are optional. They process data only if you turn them on.
| What it does | Company | Location |
|---|---|---|
| Electronic signature | Scrive | Europe |
| Slack integration, connected by you over OAuth | Slack | Per your Slack workspace |
| Website analytics — this site only, never contract data | GetResponse | Poland |
The current list, with the full description of what each one processes, is Sub-Appendix B of the DPA. We tell you before a sub-processor is added or replaced, and you can object.
The model layer
Contract text reaches an AI model because reading contract language is what the model is for. What happens to it after that is bounded.
- Your contract text is not used to train models. Lexnus does not do it, and does not permit its AI sub-processors to use your data to train or improve general-purpose or provider-owned models.
- The model only sees what the task needs. AI sub-processors are given personal data only to the extent required to provide the functionality you asked for.
- The model does not decide anything. It reads and extracts. Your published playbook returns the verdict. A rule passes or fails deterministically — the model is not consulted on the outcome.
- If you connect your own AI tool, it is yours. An external AI service you connect with your own credentials or API key is not a Lexnus sub-processor. You control it, under your own agreement with that provider, and our no-training commitment does not extend to it.
Access and authentication
| Authentication | OAuth 2.1 authorisation code flow, or a personal access token issued in Lexnus |
| Scopes | lexnous:read and lexnous:write, enforced on personal access tokens. An OAuth token carries the user's own authority, bounded by their Lexnus role. |
| Access token lifetime | 15 minutes |
| Refresh token lifetime | 7 days |
| Tenant isolation | A token reaches only the account it belongs to. Never another customer's data. |
| Session model | Stateless. Every call is authenticated independently, so there is no server-side session to hijack. |
| Document download links | Single use, 15 minutes |
| Original-file attach links | Single use, 7 days |
| Rate limit | 5 requests per second sustained per account, burst 20, on the MCP channel |
| Outbound email | Only for e-signature, and only after a human ticks a server-issued confirmation box |
The MCP reference documents the authentication model in full, including the tool-level scope for all 23 tools.
The record
Every write is recorded to an append-only log: who did it, which account and workspace, which entity, what operation, and the before-and-after values. Each row is stamped with the channel it arrived through, so a change made by an agent over MCP is distinguishable from one made by a person in the application.
Reads emit a tracing span rather than an audit row. Published playbooks are immutable — an edit creates a new version, and every analysis result stays traceable to the frozen policy state that produced it. That is what makes a verdict defensible a year later.
By plan
The infrastructure and the audit log are the same on every plan, including Free. These controls are Enterprise.
Questions
Is our contract text used to train an AI model?
No. Lexnus does not use your data to train models and does not permit its AI sub-processors to use it to train or improve general-purpose or provider-owned models. This is a commitment in the DPA, not a setting.
The one exception is an external AI service you connect yourself with your own credentials. That runs under your agreement with that provider, not ours.
Are we exposed to the US CLOUD Act?
Every essential sub-processor is a European company processing within the EEA: Clever Cloud, OVHcloud, Sweego, Better Stack and Mistral AI. No US-controlled provider processes your contract text.
Website analytics on this marketing site is the only place a standard-contractual-clauses transfer appears, and it never touches contract data.
Can an AI agent connected over MCP reach data it should not?
No. A token reaches only the account it belongs to, and an OAuth token carries no more authority than the person who issued it — bounded by their Lexnus role. Write scope is separate from read scope and enforced per tool.
Every write the agent makes lands in the audit log, stamped as arriving over MCP. If an agent changed a rule, you can see it did.
Do you have SOC 2 or ISO 27001 certification?
Enterprise plans include SOC 2 documentation and support for your security questionnaire. Ask us directly for the current status of any specific certification rather than reading it off this page — we would rather answer precisely than leave a stale claim here.
Who is the contracting entity?
Precisely AB, reg. no. 556963-5286, Nellickevägen 26, 412 63 Gothenburg, Sweden. The full terms are in the Terms of Service and the Privacy Policy.
Send us the questionnaire.
If your security team needs something this page does not answer, ask. We would rather go through it with you than publish a claim we have to qualify later.