Langdock

What it is

Langdock is a Berlin-built enterprise AI workspace: a ChatGPT-style chat, no-code agents, a workflow builder and an API, all sitting on top of whichever models you choose — current OpenAI, Anthropic, Google and Mistral models plus 40-odd others, switchable per chat or per agent. The pitch is the ChatGPT Enterprise rollout for companies whose legal team will not sign ChatGPT Enterprise. The multi-tenant cloud runs on Microsoft Azure in the EU, the company holds ISO 27001 and SOC 2 Type II, and large customers can take single-tenant, bring-your-own-cloud or on-premises Kubernetes deployments. In September 2026 Langdock dissolved its Delaware parent and re-registered as a German Societas Europaea, explicitly to answer customer questions about US Cloud Act exposure. At that point it reported a recurring-revenue run rate above €50 million, roughly 13,000 customer organizations and about 65 employees (Startup Fortune). Its only disclosed outside funding is a $3 million 2024 seed led by General Catalyst and La Famiglia.

Why it shows up in RevOps, Legal ops and Recruiting stacks

• One governed AI seat for the whole company. The buyer is usually IT or legal, not a single team: Langdock replaces the shadow ChatGPT and Claude accounts that sales, legal and talent teams were expensing. Admins decide which models exist, set usage limits per user and per agent, and export usage by user, agent and model.
• Agents reach the systems ops teams already run. The integration directory lists about 130 native connectors, including Salesforce, HubSpot, Pipedrive, Jira, Confluence, SharePoint, Google Drive, Ashby and ServiceNow, plus any remote MCP server. An account-brief agent for AEs, a clause-lookup agent for legal ops and a scorecard-summary agent for recruiters all live in one workspace, reachable from Slack, Teams or Outlook.
• It is MCP in both directions. Langdock connects to external MCP servers, and its own MCP server at api.langdock.com/mcp exposes workspace agents as tools, so Claude or Cursor users can call a governed Langdock agent without leaving their client.

Pricing reality

The Business plan is €25 per user per month billed monthly, or €20 annually, for up to 1,000 users with SSO, SCIM and model usage included. Business Max, with 5x the usage limits, is €99 monthly or €79.20 annually and can be assigned per user. Seats 51 and up get tiered volume discounts that Langdock does not publish. Enterprise (1,000+ users) is quoted.

Three meters sit outside the seat price:

• Included usage has limits. Each user gets a 5-hour session limit and a weekly limit. When either runs out, messages fall back to a cheaper default model unless an admin enables Extra Usage, billed at API rates. Enabling it applies a €1,000 monthly workspace cap by default.
• Workflows bill AI tokens separately, even on the “AI models included” plan. The free Starter tier covers 2,500 runs a month; Business workflow capacity is €539 a month for 40,000 runs or €1,199 for 100,000.
• Governance (agent review, EU AI Act and GDPR compliance templates) is free until January 1, 2027, then €2.80 per user per month annually.

For a 150-seat rollout on annual billing with 10 Business Max power users and no workflows, budget about €43,000 a year before volume discounts (estimate: 140 × €240 plus 10 × €950). Add €6,500 a year once workflows go past the Starter tier.

Best for

IT, legal ops or RevOps leaders at EU-headquartered or EU-regulated companies of 50 to 1,000 employees who need a company-wide, model-agnostic assistant with agents, and whose procurement has rejected US-owned vendors over data access.

Do not buy it if you are a Microsoft 365 E5 shop happy inside Microsoft’s EU Data Boundary — Copilot is already in the tenant. And do not buy it if the real job is enterprise search across dozens of permissioned sources; that is a Glean problem.

Versus the alternatives

ChatGPT Enterprise is the category’s installed-base leader. Pick it when a US-controlled vendor is acceptable and you want OpenAI’s newest features first. Pick Langdock when legal needs an EU-domiciled contracting party or you want Claude and Gemini beside GPT under one bill.

Microsoft 365 Copilot, at $30 per user per month, is the other share leader. Pick it when work lives in Outlook, Teams and SharePoint. Pick Langdock when model choice and CRM or ATS agents matter more than Office integration.

Dust is the fastest-growing European rival, raising a $40 million Series B in May 2026. Pick Dust when agents that take action across your SaaS stack are the main job. Pick Langdock when the main job is a company-wide chat seat with agents on the side, at €20 against Dust’s €29 Pro seat.

Mistral Le Chat Enterprise is the pick when every layer, model maker included, must be European. Langdock routes most traffic to US model makers.

If none fit because data can never leave your network, look at self-hosted Onyx or Langdock’s own on-premises option, which starts at 5,000 seats.

Watch-outs

• The SE move fixes the corporate layer, not the infrastructure layer. Multi-tenant Langdock runs on Microsoft Azure, and most models come from US providers, which remain subject to the Cloud Act. Guard: have legal read the subprocessor list, and if that exposure is disqualifying, price bring-your-own-key routing to an EU provider such as Mistral, or a dedicated deployment, before signing.
• Included usage runs out quietly. Heavy users drop to the fallback model mid-week and report that “Langdock got worse.” Guard: watch the usage export for the first 30 days, move the top 5–10% of users to Business Max, and set per-user Extra Usage caps rather than one workspace cap.
• Workflow token costs arrive on a separate invoice line. An agent-node workflow scheduled across your pipeline can spend more than the seats. Guard: set workspace and per-workflow spend limits before publishing any scheduled workflow, and budget workflow tokens as their own line.
• BYOK is a setup-time decision. You choose it when upgrading from trial; switching later needs Langdock support, and a BYOK workspace must configure its own completion, embedding and image models. Guard: decide model ownership during the trial, not after rollout.