Claude for Work vs Langdock

Claude for Work

The commercial terms state that Anthropic may not train models on Customer Content, a data processing agreement is accepted through those same terms, and Enterprise admins can set custom retention for conversations, projects and artifacts with every change tracked in audit logs and flagged-content retention spelled out in writing — most of what a works council needs on paper. The default is indefinite retention, however, and the retention controls exclude Claude Managed Agents and any feature running Claude Code in the cloud, exactly the agent surfaces where staff activity would accumulate.

Langdock

Training exclusion is written into the contract, not just marketing: the terms state Langdock engages model providers only under agreements barring them from training on customer content, and for the Azure models abuse-monitoring storage and human review by Microsoft are disabled under zero data retention. Admins set the chat retention period themselves, file retention is documented per storage location, a data processing agreement is concluded automatically with 30-day post-termination deletion, and ISO 27001 and SOC 2 Type II reports are published. We found no public information on a deletion audit trail or on abuse-monitoring terms with the other model providers, which keeps this short of a full account.

Claude for Work

The model families are named and the privacy policy states plainly that personal data is transferred to servers in the US or to other countries outside the EEA and UK, so the processing location is documented but not European, and we found no public information on an EU processing option for the hosted service. For the Amazon Bedrock route, the page defers to AWS's own user guide and lists no regions itself.

Langdock

EU-only processing is the default and leaving the EU requires an active customer choice, with hosting in Frankfurt on Microsoft Azure under the EU Data Boundary and data at rest staying in EU data centres even for global models — commitments that sit in the data processing agreement, not only on a marketing page. Bring-your-own-key is documented so the workspace runs only on customer-configured models, and own-cloud or on-premises deployment is offered from 5,000 seats. We found no public information stating the inference region for each of the forty-plus available models individually.

Claude for Work

Enterprise plans carry single sign-on with domain capture, SCIM provisioning, audit logs and usage analytics, and the role-based permissions documentation is unusually thorough: groups synced from the identity provider, per-role model access, connector permission levels that default to "Needs approval", per-group spend limits, and retrieval through the Pipedrive connector that respects existing user permissions. For a works-council agreement I would still need usage reporting that can be aggregated or pseudonymised, and we found no public information on either that or on streaming audit logs to a SIEM.

Langdock

SAML 2.0 with any compatible identity provider, SCIM on the business tier, role-based access control, mirrored source-tool permissions and a governance module where admins review, approve, flag or disable agents against automated compliance rules that check configuration rather than what people say in chats — and usage analytics can be switched off per workspace, with usage exports by CSV and API. For my side of the table the gaps matter: we found no public information on exportable audit logs, data-loss prevention, prompt-logging rules set out in writing, or usage reporting that can be aggregated or pseudonymised for a works-council agreement.

Claude for Work

Connectors reach Gmail, Google Drive, Slack and Microsoft 365, a marketplace offers MCP connectors such as a Pipedrive integration with read and write tools over OAuth, and the Claude API plus webhooks and external APIs are documented. We found no public information on sync speed or completeness, on building custom connectors, or on how the assistant cites the sources behind an answer.

Langdock

MCP support is unusually deep: a curated directory of 79 officially operated remote servers, custom servers connectable with OAuth or API keys, sandboxed interactive apps with confirmation gates, the A2A protocol, native connectors for document management, email, communication tools and ticketing, plus Zapier reach into over 7,000 apps, Excel and Outlook add-ins, mobile apps and a unified model API. We found no public information on source citations for grounded answers, on sync frequency or completeness, or on exporting agents and configuration when a customer leaves.

Claude for Work

Managed agents run multi-step workflows that act on connected systems under scoped permissions, and the approval machinery is what a works council wants on paper: a newly created role defaults to "Needs approval" on every connector, and letting members set "Always allow" on write-capable connector tools is off by default at the organization level. Limits, failure handling and even the fact that there is no approval step before a run starts are documented candidly, but we found no public information on testing or evaluation before an agent is released, on versioning, or on monitoring cost per agent.

Langdock

This is where my checklist is largely satisfied: workflow action nodes run only after human confirmation, approvals can be conditional so only some runs need a person, agents publish in versions with edits held as drafts, and governance records a pass or fail on every new agent version while showing each agent's instructions, integrations, costs and history. A build-and-test toggle, per-agent cost and analytics monitoring, and a published statement that AI output may be inaccurate and is not warranted give documented limits. We found no public information on evaluation against test sets or on attributing every run to the agent and the person who deployed it.

Claude for Work

EEA, Swiss and UK customers contract with Anthropic Ireland Limited, a subprocessor list is published, and standard contractual clauses are the stated transfer safeguard — but the model maker and parent is Anthropic PBC in the US, and the privacy policy says personal data goes to US servers or other countries outside the EEA and UK. We found no public information on EU hosting or EU inference for the product itself.

Langdock

The contracting entity is Langdock GmbH of Berlin, registered at the Charlottenburg local court, hosting runs on Microsoft Azure in Frankfurt, and the subprocessor list is public with locations — OpenAI Ireland, Google Cloud, AWS and Azure all shown as EU, Sentry's error data processed in the EU under the Data Privacy Framework and standard contractual clauses, and any third-country model processing only on explicit customer instruction. A data processing agreement with export and 30-day deletion terms plus published ISO 27001 and SOC 2 Type II reports round out the picture. We found no public information on the ownership structure, which alone keeps this below the top level.

Claude for Work

Seat prices are public — $0 Free, "$17 Per month with annual subscription discount ($200 billed up front). $20 if billed monthly." for Pro, Max "From $100 Per month", and "$20 per seat per month plus usage billed at API rates" for Enterprise with a 20-seat minimum and annual commitment — and agent runs are clarified as carrying no price of their own, billed at each model's rates. We found no public information on the Team seat price, on the usage limit figures, or on the API rates themselves, so a buyer cannot compute a full annual cost from these pages alone.

Langdock

Seat prices are public — €23.20 per user per month standard and €99.00 Business Max with 5x more usage than Standard seat, save 20% on annual billing — alongside workflows at €449.00 per workspace per month for 40k workflow runs, a governance add-on at €2.80 per user per month free until January 1, 2027, per-token API rates, deployment minimums by seat count, and contract notice periods and renewal pricing in the published terms. What a buyer still cannot compute: the fair-use quantum behind a seat, overage rates ("contact us for higher limits"), and the Enterprise tier's custom price.