Langdock vs nele.ai

Langdock

The terms commit that neither Langdock nor any model provider it engages may train on customer content, Azure models run with modified abuse monitoring described as zero data retention for prompts and outputs, and a DPA under Art. 28 GDPR is automatically concluded. Admins set the chat retention period, deletion behaviour is documented per feature for chats, agents, folders and knowledge bases, and data is deleted within 30 days of termination. I would want the same abuse-monitoring and zero-retention detail for every provider, not just Microsoft, before calling it fully accounted.

nele.ai

The privacy policy instructs the AI providers that user inputs may not be stored or used for training — framed as subject to the functions the providers make available — and states that logfiles are deleted or anonymised after at most 30 days and that personal data is deleted once its purpose ends; the vendor also claims GDPR conformity and ISO 27001 certification. Retention is stated for logs only, and beyond an agreement statement for the email service Brevo we found no public information on retention periods for chats and files, zero-retention terms with the model providers, or the terms of a data processing agreement covering the assistant itself.

Langdock

EU-only processing is the default and inference leaves the EU only if a customer actively selects a global-deployment model, with data at rest still held in EU data centres and Frankfurt hosting under the Azure EU Data Boundary documented. Forty-plus models are offered across named providers, bring-your-own-key locks a workspace to only customer-configured models, and single-tenant, own-cloud and on-prem (Kubernetes via Helm) deployments are documented from 2,000 to 5,000 seats. A published per-model inference-region table and open-weight or self-hosted model support are not documented on the captured pages.

nele.ai

A German server location is published, hosting subprocessors are named as Hetzner in Germany and Microsoft Azure in Frankfurt, and the policy states that user inputs are processed on the vendor's own servers rather than sent directly to the AI providers. We found no public information on which models sit behind the assistant, who runs them, or on any administrator option to pin or restrict processing — even in the captured pages dedicated to AI models and their providers.

Langdock

SAML 2.0 SSO with any compatible identity provider, SCIM, signed authn requests, domain verification, enforced MFA at the identity provider and IP restrictions give IT real control, and retrieval mirrors source-tool permissions one to one so an agent cannot read beyond its user. Usage analytics can be enabled or disabled per workspace with CSV and API exports of cost and token data. The governance module lets admins review, approve, flag or disable agents with compliance rules re-checked on every published version; we found no public information on exportable audit logs or streaming to a SIEM.

nele.ai

The captured pages include a product function page on single sign-on with SAML 2.0 and an administrator section covering knowledge bases, and the privacy policy describes user account registration. We found no public information on roles and groups, SCIM provisioning, group-level permissions over models and knowledge sources, audit logs or usage reporting — exactly the controls I need before an assistant is let anywhere near company systems.

Langdock

MCP is first-class: a directory of 79 officially operated remote servers, custom MCP servers connectable with OAuth 2.0 or API-key auth, MCP tools surfaced as actions with confirmation flows, MCP Apps rendering interactive UI in a sandboxed iframe in chat, an agent-to-agent protocol, plus a unified API, native connectors, Office add-ins, desktop and mobile apps and Zapier reach into 7,000 apps. We found no public information on incremental sync behaviour, webhooks, or source citations on grounded answers, which I need before pointing this at ERP and CRM flows.

nele.ai

Function pages on knowledge bases and retrieval-augmented generation, including an administrator view, are part of the capture, so grounding answers in company knowledge is clearly part of the design. We found no public information on connectors to file stores, wikis, mail, CRM, ticketing or ERP, on sync behaviour, on an API or webhooks, on open protocols such as MCP, or on how answers cite their sources — the pieces I would need before wiring anything into ERP, CRM or the wiki.

Langdock

This is what I look for: action nodes with human confirmation so nothing is sent without approval, workflow pauses with approve, reject and form input, conditions so only some runs need a person, a workflow monitor for pending steps, and a builder test toggle with test runs that skip notifications. Agents get versioned publishing where later edits stay drafts until published, scheduling, scoped write folders, subagents, Slack and Teams channels, and a governance view exposing each agent's instructions, integrations, costs and history with the power to disable it until fixed. Evaluation against defined test sets and documented failure handling beyond a general output disclaimer are not documented.

nele.ai

We found no public information on agent capabilities — multi-step tasks, tool use or actions in other systems, approval steps before consequential actions, run logs, or testing and evaluation before release. The homepage, the data protection and security page and the terms of use were among the pages we read.

Langdock

Langdock GmbH in Berlin (registered under HRB 254499) is the contracting entity under German law with exclusive jurisdiction in Berlin, hosting runs on Microsoft Azure in Frankfurt with the EU Data Boundary, and a published subprocessor list names every model provider with its location, including OpenAI Ireland, AWS, Google Cloud and Azure, plus Sentry in the US covered by SCCs and Data Privacy Framework safeguards. Transfers outside the EU happen only on the customer's instruction, and ISO 27001:2022 and SOC 2 Type II certificates are published. We found no public information on the company's ownership structure, so I credit the dimensions the evidence proves.

nele.ai

The contracting entity is GAL Digital GmbH of Hungen, Germany, with register court, register number and managing directors published, hosting is stated as German on named infrastructure — Hetzner and Microsoft Azure in Frankfurt — and the privacy policy lists subprocessors with their locations, almost all German, with Stripe and Webflow in the USA under the Data Privacy Framework and standard contractual clauses. For an AI product the decisive subprocessor is the model provider, and we found no public information naming the providers behind the assistant or where their processing runs. The vendor claims GDPR conformity and ISO 27001 certification, and we found no public information on a data processing agreement covering the assistant itself.

Langdock

Public per-seat prices of €23.20 standard and €99.00 Business Max with 5x usage (both excluding VAT, 20% off annual), workflow tiers with 2,500 runs included and €449 per workspace per month for 40,000 runs, a €2.80 governance add-on free until January 1, 2027, and per-token API rates make most of an annual bill computable. The included usage quantity per seat rests on a fair-use policy rather than published numbers, overage is not priced, the exact governance add-on price varies by seat count without a published scale, and Enterprise pricing is on request.

nele.ai

The published statement is that one license covers all users — a cost unit, but no figures. We found no public information on what the license costs, which models or functions it includes, usage limits, overage, minimum seats or term, so an annual cost cannot be worked out from the pages we read.