One safe place for your whole company to use AI. Sensitive data is masked before it leaves your company, every use is audited, and you choose which data may be shared. Runs on your servers or in our secure cloud.
Contract review, clause comparison, summaries of long filings and first-draft correspondence — inside a workspace per matter that nobody outside the matter can see.
Summarize notes, draft discharge letters and referrals, look up guidelines — with patient identifiers pseudonymized before anything reaches the model.
Developers get AI coding agents that do real work — inside an isolated sandbox per session, with source, keys and production out of reach.
Analyze spreadsheets, draft board reports, answer policy questions — with confidential figures classified and every export traceable.
Built for regulated environments
🏠Self-hosted · on-prem or your cloud 🪪SSO (OIDC) + MFA 🛡️LLM firewall on every request 📋Full audit trail 🔑Bring your own AI provider 🔍DLP for AI · sensitive data detected & maskedYour people already use AI. The question is whether it happens in fifty personal accounts you can't see — or in one place you control.
Employees paste contracts, patient notes and source code into whatever chatbot is fastest, under personal logins, with no record of what left.
Everyone signs in with company SSO to a workspace for their team. Every request passes through the firewall; every flow of data is recorded. The platform runs on your servers.
Prompts, attachments and tool results are checked against your policies before they leave your network, and model responses are checked on the way back.
Plug in your SSO, create workspaces for teams, and define firewall policies: what is personal data, what is confidential, what may never leave.
Lawyers, clinicians, engineers and analysts use one familiar AI tool. Each request is checked and masked by the firewall on the way to the model you chose.
Your security team gets the audit trail — who, when, which policy, what was masked — and exports for compliance. Content stays with its owner.
Instead of fifty personal accounts, one workspace per team behind company SSO — with roles, MFA and access that ends the day someone leaves.
The LLM firewall checks each request and response against your policies. The audit log records every flow of company data — and is built for auditors, not just engineers.
On-prem or in your cloud, with the AI provider of your choice. Even the platform admin can't read anyone's conversations or documents.
The same capabilities your people already want — inside your perimeter.
| public AI chat · personal accounts | ARX RUN | |
|---|---|---|
| Where data goes | Straight to a third party, unmasked | Through your LLM firewall: personal data, secrets and classified figures masked or withheld first |
| Visibility & audit | None — you don't know what left | Full audit trail of every request: who, when, workspace, policy, what was masked; exportable for compliance |
| Access control | Personal logins, no offboarding | Company SSO with MFA, role-based workspaces per team, access revoked with the identity |
| Deployment | SaaS, their terms | Self-hosted on-prem or in your cloud, with the AI provider you choose |
| Who sees content | The provider, and anyone with the login | Only its owner — not even your platform admin |
*structural properties of the platform, not benchmark results. Each session gets its own isolated worker; every user connects their own provider key; every request and connector call passes through the firewall and is written to the audit log.
No. It is the layer between your people and the AI models you allow. You bring your own provider account (today Claude; more coming). ARX RUN adds identity, workspaces, the LLM firewall, the audit trail and sandboxed agents around it.
From the user's workspace through the firewall, which masks or withholds sensitive data according to your policies, then to the provider you chose. The audit log records the flow — who, when, which policy, what was masked — but not the content.
Those are AI features inside one vendor's app. ARX RUN covers work that spans systems, data those tools cannot reach (DMS, hospital systems, ERP, file shares), autonomous agents, self-hosted deployment and full auditability. It does not replace the AI button inside Outlook — and doesn't try to.
No. Operators see state and counts — sessions running, policy hits, projects — never the content of sessions or documents. The boundary is enforced by the server, not by policy.
You can. Self-hosting on your own servers (podman compose) or cluster (Helm) is the primary model. A secure cloud option exists for teams that prefer not to operate it.
Claude, through the official Claude Agent SDK. Codex, Cursor and OpenCode are planned. You decide which providers your company allows; all of them get the same firewall, audit and workspaces.
ARX RUN sits between your people and the model. You decide which providers your company allows — and every one of them gets the same firewall, the same audit trail and the same workspaces.
C
ClaudeAnthropic
available
C
CodexOpenAI
coming soon
C
CursorAnysphere
coming soon
O
OpenCodeopen source
coming soon
See the firewall, the audit trail and the workspaces on your own data.