Questions worth asking.

airlock is one connector that links every AI client your team uses (Claude, ChatGPT, Cursor, Copilot, Gemini, Claude Code) to any MCP server or OpenAPI spec. Set policy once and audit everywhere. Skills and Agents become versioned, verified objects shared across the org.

Microsoft Agent 365 governs the Microsoft tenant; airlock governs across vendors (Claude, OpenAI, Cursor, Copilot, Gemini, anything OpenAPI).

ServiceNow AI Control Tower is enterprise-scale with Now-Platform lock-in; airlock is mid-market priced (€19/user/month) and vendor-neutral.

Langdock / CompanyGPT is a single managed AI platform for employees; airlock is the connector across the AI clients your team already uses.

OSS MCP gateways (Kong, MintMCP) are technical plumbing without Agents-and-Skills-as-objects.

The Model Context Protocol, an open standard for AI clients (Claude, ChatGPT, Cursor, etc.) to talk to external tools and data sources. Donated to the Linux Foundation in December 2025 under the Agentic AI Foundation. airlock is an MCP server that proxies and governs every other MCP in your org.

A Skill is a packaged prompt or workflow (versioned, verified) that any AI client can use through airlock. An Agent is a composable object built from Skills + tool access, with fork-and-lineage so you can branch one for a client engagement and trace every version's origin.

Proxy. airlock sits in front of your existing MCPs and APIs. Credentials stay server-side in a vault; agents never see your real API keys. You can swap out any backend without re-onboarding clients.

The core use case. Put one connector in front of every AI client your team uses (Claude, ChatGPT, Cursor, Copilot, Gemini, Claude Code) and govern them in one place. Set policy once and it applies everywhere: approvals on destructive actions, a full audit log, a kill switch, role-based access, and credentials kept server-side in a vault so agents never see your real keys. IT and security get a single source of truth for what every person and agent can do, across vendors.

Consultants and agencies can collaborate with their customers through airlock, sharing connectors, Skills, and Agents while keeping access under control. You decide exactly who can use what, revoke access at any time, and keep an audit trail of every cross-org interaction. Re-use a canonical Skill across clients with fork-and-lineage, without leaking one client's setup into another's. A separate scope per engagement is an Enterprise engagement, scoped and priced on demand. See Agencies & consultancies for the details.

Running a training, workshop, or bootcamp? Stand up a temporary environment of Skills, Agents, and connectors so participants work in a governed sandbox: the right tools, the right guardrails, nothing they shouldn't touch. When the session ends, tear it down or revoke access in one move. No leftover credentials, no lingering access, no manual clean-up across a dozen accounts.

Free for solo builders (1 user). Teams is €19 per user / month on a yearly plan, or €24 month-to-month, with unlimited everything. Agents you run on an agent runtime (AWS Bedrock AgentCore, Vercel Agents, Azure AI Foundry, and others) are an add-on at €5 per running agent / month on paid plans. Enterprise is contact sales. Full breakdown on the pricing page.

1 user, unlimited MCPs, Agents & Skills, personal memory, audit log (7-day window), community Slack support. No credit card.

Yes. Teams starts with a 15-day free trial, no credit card required. Solo is free forever (1 user).

No. airlock is priced per user per month. We never markup tokens, and we don't see them. Your LLM bills stay between you and Anthropic / OpenAI / etc.

The data you put into airlock is stored at rest in the EU: AWS Frankfurt (eu-central-1). Sovereignty is a hard filter for us, and there's no US storage fallback. What we won't claim is that nothing ever leaves the EU. Our pages are served through a global CDN edge, browser analytics reach PostHog through an edge proxy before landing in its EU project, AWS requires our TLS certificates to be issued in us-east-1, and a handful of website and account tools we use sit in the US. Separately, if you supply your own credential for a non-EU model provider, requests on that route leave the EEA because you configured them to; we never enable that for you. Sections 5 and 8 of the privacy policy name every one of them.

GDPR-aligned by construction. ISO 27001 certification is in progress. EU AI Act obligations (logging, human oversight, traceability) are produced by default. Every tool call logged, every approval attributed, every Agent and Skill versioned.

No, unless you ask us to. Audit logs, approval payloads, and proxied request/response data are for your auditing. We only access them when you open a support ticket asking us to investigate, or where strictly required by law.

We notify affected customers within 72 hours per GDPR Article 33. Our threat model and response runbook are designed around the proxy-as-defence pattern: credentials stay encrypted at rest in the vault and are only decrypted at the moment of API execution, so a breach of the cache exposes the least data possible.

No. airlock doesn't train models, and we don't pass your data to any third party for training. Your prompts, tool calls and Skills are yours; we proxy them, we don't reuse them.

First connector live in under 60 seconds for solo signup. For a team rollout: 30-min setup call with a founder, your first MCP set up for you, policy rules drafted. Most teams ship their first governed flow on day one.

Drop an OpenAPI spec into airlock. We generate the MCP server, OAuth-aware, governed by your policy. No engineering required. An ops manager or AI practice lead can stand up a new integration before lunch.

On demand. The default is SaaS (EU-hosted on AWS Frankfurt), but dedicated and self-hosted deployments are available on the Enterprise tier. Email sales@air-lock.ai.

Yes. Every audit event is SIEM-exportable (JSON / CEF) from day one. Push to Splunk, Datadog, Elastic, your S3 bucket, wherever you keep your security telemetry.

Yes. White-label and multi-tenant deployments are available on the Enterprise tier. Email sales@air-lock.ai to discuss.

Org defaults flow down, and each level below adds to them. Where a person sits in more than one, the most permissive right wins. The exception is a rule the organization marks locked: nothing below it can widen that. A separate scope per client engagement, with its own MCPs, Skills, policy and audit log, is an Enterprise engagement, scoped and priced on demand.

Yes. Fork-and-lineage: the canonical Skill stays in the org library; each client engagement gets its own version trail. Roll back per-client without affecting the shared baseline.

In design. If you sell AI delivery work to clients and airlock would make your delivery more governable, we'd love to talk early. Drop us a line at sales@air-lock.ai.

Yes. Per-user pricing scoped per workspace plus per-project usage reports. Cost rolls up cleanly to engagement P&L and matches how consultancies bill clients.

Email info@air-lock.ai or DM the founders on LinkedIn. Direct line to the founders: answers in hours, not tickets.