§ ENT · Security Enterprise · Security

Where your
data goes.

Transparency instead of "enterprise-grade security" banners. Below is the concrete detail: what data we see, where it lives, what reaches the models, what goes to logs, and which safeguards are built into the agent by default. No "bank-level" phrasing, because it means nothing.

EMAIL SALES JOIN THE WAITLIST
§ ENT · S1 Where data lives

What we see, what we store, what we don’t.

A clear picture of the data routes — from the client to Anthropic and back. Storage, retention and access included.

01 · Relay

Routed through our relay

All LLM requests pass through it and history is stored there (EU, Stockholm).

02 · History

Transcripts in the relay DB

Sessions and messages are written to Postgres: role, content, tool calls, timestamps. Needed to "return to an old chat". On request we purge, export or migrate them to your self-host.

03 · LLM providers

Anthropic and compatible

Chat traffic goes to Claude via the API. Per their policy, API requests are not used to train models. The relay can fail over between several backup providers — on request we pin it to one.

04 · Permission engine

Protection from accidental deletes

Every tool operation is classified: destructive bash patterns (rm -rf, git reset --hard), dangerous PowerShell cmdlets, an SSRF guard for fetch, a read-only pool for safe commands. YOLO auto-approves only what is safe.

05 · Local

No whole-repo upload

In Code mode the agent runs on your machine and doesn't index or bulk-upload your repository — there's no background copy of your codebase in the cloud. Only the specific prompt text and tool results the agent needs are sent to the model provider.

06 · Deletion

Right to be forgotten

The account and all data are deleted on request — sessions, messages, the MCP token cache, api_calls. The process is spelled out in the contract. You initiate it; turnaround is one business day.

§ ENT · S2 Built in vs. on request

Technical facts, not marketing phrasing.

Built into the product

What works right now

  • TLS on every connection client ↔ relay ↔ LLM provider
  • Postgres with encryption at the hosting disk level
  • Permission engine: destructive patterns, dangerous PowerShell cmdlets, SSRF guard
  • Audit table api_calls with every request: user_id, model, tokens, cost
  • JWT with expiry + revoke on logout; HttpOnly session cookies
  • Rate limiting on auth endpoints against brute force
  • Anthropic API: no training on your data (per their policy)
Scoped under contract

For compliance-heavy teams

  • Self-hosted relay on your infra (Kubernetes or docker-compose)
  • Your own LLM provider API keys — traffic never passes through our pool
  • Logs to your SIEM / ELK, not only our DB
  • A single LLM provider instead of auto-routing (when knowing where matters)
  • NDA, DPA, a confidentiality clause in the contract
  • History disabled in our DB — everything stays only with you
  • Pen-test report from SECURITY_AUDIT.md (available after NDA)

Still have questions?

If you need detail on a specific point — a data route, encryption, NDA, compliance — write to us and we’ll answer technically, without generalities.

sales@agentnine.io JOIN THE WAITLIST