Langdock vs nele.ai

Langdock

No-training is contractual, not just marketing: the terms state Langdock engages model providers only on agreements where those providers are "also prohibited from using Customer Content to develop, train or improve their AI Models", the DPA is auto-concluded, admins set a retention period with automatic chat deletion, per-artifact deletion behavior is documented, data is deleted 30 days after termination, and ISO 27001 plus SOC 2 Type II certifications are published. Zero-data-retention abuse monitoring is documented for the Azure-hosted models specifically; we found no public information on abuse-monitoring terms for the other model providers.

nele.ai

The privacy policy says user inputs are processed on the company's own servers and passed to AI providers only with the instruction that they may be neither stored there nor used for learning, qualified as subject to the functions the providers make available. Logfiles are kept for at most 30 days and personal data is deleted once its purpose or legal basis lapses, but I found no public information on retention periods for chats and files, zero-data-retention terms, or the terms of a data processing agreement for the assistant itself — the only processing agreement described covers the email vendor Brevo.

Langdock

This is the model story I want for a regulated rollout: model-agnostic access to 40+ models, EU-only processing by default with global inference only when a customer actively selects such models, third-country transfer only on controller instruction written into the DPA, and bring-your-own-key plus dedicated own-cloud and on-premises deployments. A published per-model inference-region list and open-weight self-hosted models we found no public information on.

nele.ai

A German server location is stated, Microsoft Azure in Frankfurt and Hetzner in Germany are named for infrastructure, and the privacy policy says user inputs are processed on the vendor's own servers rather than sent directly to AI providers. Captured pages are dedicated to the AI models and their providers, but they confirm no model names, providers or inference regions to me, and I found no public information on pinning processing to the EU, excluding providers, or bringing my own key — the exact questions I would need answered before eight hundred staff use it.

Langdock

SAML 2.0 with any identity provider, signed requests and domain verification, SCIM named on the Business plan, role-based access control with custom roles and group grants, IP restrictions, usage analytics togglable per workspace with CSV and API export, a governance console to review, approve or disable agents with compliance checks on every new version, and mirrored permissions that reflect source-tool access 1:1 — which answers my SharePoint concern directly. We found no public information on an exportable or SIEM-streamable audit log, which my SOC will want before eight hundred seats.

nele.ai

A product page documents single sign-on with SAML 2.0 and another is written for administrators, which is where my interest starts for a rollout of this size. Beyond that I found no public information on SCIM provisioning, group-based permissions for models, agents and connectors, retrieval that inherits source-system permissions such as SharePoint's, or audit logs I could stream to the SIEM.

Langdock

Open protocols are first-class: a directory of 79 verified MCP servers, custom MCP servers with several OAuth types, discovered tools converted into actions with confirmation flows, an MCP Apps SDK across major frameworks, and A2A agent communication, alongside a unified API, Outlook and Excel add-ins, Zapier reach, and agents usable in Slack and Teams. We found no public information on incremental connector sync or source citations on grounded answers.

nele.ai

The captured pages document knowledge bases with retrieval-augmented generation, so a company knowledge layer exists, but I found no public information on connectors to file stores, wikis, mail or ticketing, on sync behaviour, an API, or how answers cite their sources. Nothing here shows how the assistant would respect the permissions of a source system, which is non-negotiable for us.

Langdock

Agents act through integration actions and workflows, with human confirmation on action nodes, approve/reject and form input nodes that pause runs, scheduling outside the builder, publish-and-draft versioning with a build-and-test mode, and per-agent instructions, analytics, costs and history visible in governance with automated checks on every published version. Fair-use limits and a stated no-warranty position on AI outputs are documented; evaluation against defined test sets before release we found no public information on.

nele.ai

We found no public information on agent capabilities. No captured page documents multi-step tasks, actions in other systems, approvals, run logs, testing or stated limits, and the product's category label alone is not capability evidence I can score.

Langdock

Langdock GmbH registered in Berlin, exclusive jurisdiction Berlin, hosting in Frankfurt on Microsoft Azure with the EU Data Boundary, a published subprocessor list naming AWS, Azure, Google Cloud and OpenAI Ireland with locations, Sentry for error tracking under Data Privacy Framework and standard contractual clauses, EU-only inference by default with third-country transfer only on customer instruction, and an auto-concluded DPA with export and 30-day deletion documented. We found no public information on the ownership or parent-company situation, which keeps this short of the top level.

nele.ai

The contracting entity is GAL Digital GmbH of Hungen, Germany (Amtsgericht Gießen, HRB 8432), a German server location is stated with Microsoft Azure in Frankfurt and Hetzner named for infrastructure, and the privacy policy lists subprocessors with locations — mostly German, plus Stripe and Webflow in the USA backed by the Data Privacy Framework and standard contractual clauses. For an AI product the decisive subprocessor is the model provider, and here the policy refers to AI providers while its subprocessor list covers hosting, email, cookie consent, shop data, payments and the website only, so where model processing happens is unconfirmed to me. Ownership is not documented in what I hold, so I score what is proven: a German entity on named German infrastructure with listed subprocessors, held short of full marks by the unnamed model layer.

Langdock

For my seat count I can get most of the way there: "€23.20 Standard seat" and "€99.00 Business Max seat" per user per month excluding VAT with "5x more usage than Standard seat", governance at "€ 2.80 Per user / month (excl. VAT)" free until January 1, 2027, workflows from an included 2,500 runs to "€ 449.00 Per workspace / month (excl. VAT)" with 40,000 runs, per-token API prices, bring-your-own-key where you "pay Langdock only for the platform license", a 20% annual saving, and contract terms with termination notices and pro-rata mid-term additions. Enterprise is "Custom. Contact us", exact fair-use figures live in a separate policy, and we found no public information on overage rates, so the full annual cost is not computable from the pricing pages alone.

nele.ai

A pricing page is published and the licensing model is stated as "One license for all users", but I found no public information confirming price figures, per-seat or usage costs, tier inclusions, usage limits or the pricing of enterprise features such as single sign-on. I cannot compute an annual cost for eight hundred seats from what is captured.