Every European AI agent platform reviewed
Berlin, Germany
Business €25 per Standard seat or €99 per Business Max seat per user per month excluding VAT, 20% off yearly / Workflows add-on from €539 per workspace per month / Enterprise on request
7-day free trial with €5 of AI model credits, no credit card required
Best for: Companies that want a model-agnostic chat with custom agents and a named EU host
Langdock is a Berlin company whose platform combines a chat, custom agents, workflows, integrations and an API, with the chat described as model agnostic. Its imprint names Langdock GmbH at Greifswalder Str. 212, 10405 Berlin, registered at the Amtsgericht Berlin (Charlottenburg) under HRB 254499 with VAT ID DE361952314 and Jonas Beisswenger, Tobias Kemkes and Lennard Schmidt as managing directors.
The security page states that the multi-tenant SaaS is hosted via Microsoft Azure on servers inside the EU, that the application and most models, configurable, are hosted entirely in the EU, and that customer data is never used to train models. It claims ISO 27001 and SOC 2 Type II. For large customers it offers a dedicated single-tenant deployment, bring-your-own-cloud from 2,000 seats and on-premises on Kubernetes from 5,000 seats.
Pricing is published: the Business plan is €25 per Standard seat or €99 per Business Max seat per user per month excluding VAT, with 20% off when billed annually. Workflows are an add-on, with 2,500 runs a month in the Starter package and €539 a month for 40,000 runs, and model usage in workflows is billed at API pricing. A 7-day trial with €5 of model credits needs no card.
What Langdock does well
- Per-seat prices published in euros, including the higher-usage seat
- Multi-tenant SaaS on Azure inside the EU, per the security page
- ISO 27001 and SOC 2 Type II claimed
- Own-cloud and on-premises deployment for large customers
- Bring-your-own API keys and a unified API across model providers
Where Langdock falls short
- Hosted on Microsoft Azure, a US provider, even though the servers are in the EU
- Most models, not all, are hosted in the EU; the rest depends on configuration
- Own-cloud and on-premises only from 2,000 and 5,000 seats
- Workflows are a separate paid add-on with run limits
Standout feature. It is the only platform in this category that publishes a seat price, names an EU cloud region type and claims both ISO 27001 and SOC 2 Type II on one page.
Paris, France
Free with 500 credits, lifetime / Pro €24 per seat per month with 8,000 credits, billed yearly / Max €120 per seat per month with 40,000 credits / Enterprise on request
Free plan with 500 credits
Best for: Teams building agents on top of many company data sources with a published price
Dust is a Paris platform for custom agents. Its pricing page lists custom agents with skills, knowledge and tools, multi-agent orchestration on schedules and triggers, native and remote MCP servers, connectors to more than 20 data sources and more than 20 models from OpenAI, Anthropic, Google, Mistral and DeepSeek.
The privacy policy names Permutation Labs at 86 avenue de Wagram, 75017 Paris as the data controller and says it uses Google Cloud infrastructure and QDrant Cloud for embeddings.
The pricing and security pages state EU or US data residency, SOC 2 Type II, GDPR compliance, AES-256 at rest, no model training and zero data retention at third-party model providers. The terms page redirects to a Notion document that could not be read, so the contracting entity is taken from the privacy policy only.
Pricing is published: Free with 500 credits for life, Pro at €24 per seat per month billed yearly with 8,000 credits, Max at €120 with 40,000 credits, and Enterprise on request with single-tenant deployment, SCIM and audit logs. Programmatic usage is $0.01 per credit.
What Dust does well
- EU or US data residency selectable on every plan
- More than 20 models and 20 data connectors on the Business plans
- Free plan and published euro prices
- SOC 2 Type II, SSO and, on Enterprise, SCIM and audit logs
- Single-tenant deployment on Enterprise
Where Dust falls short
- Runs on Google Cloud infrastructure, a US provider
- Model calls go to US providers such as OpenAI, Anthropic and Google, with the stated zero-retention commitment
- Usage is metered in credits, which makes monthly cost less predictable than a flat seat
- The terms are hosted on a Notion page, so the contracting entity is stated only in the privacy policy
Standout feature. It lets a customer choose EU or US data residency on every plan, including the free one.
Berlin, Germany
Community Edition free to self-host / Starter €20, Pro €50, Business €667 per month billed annually, by executions / Enterprise on request
Free self-hosted Community Edition
Best for: Engineering teams that want agents as steps in a workflow they can also self-host
n8n is a workflow platform whose site lists building AI agents and RAG among its use cases. The imprint names n8n GmbH at Novalisstr. 10, 10115 Berlin, registered at the Amtsgericht Charlottenburg under HRB 212509 B with VAT ID DE327574983 and Jan Oberhauser as managing director. The legal page also lists AI Terms, Credit Terms, a Data Processing Agreement, a security page and a Trust Center.
The pricing page states that cloud plans are hosted in Frankfurt, Germany. Plans are priced by monthly workflow executions rather than users or steps: Starter at €20 a month for 2,500 executions, Pro at €50 for 10,000, Business at €667 for 40,000 and self-hosted only, and Enterprise on request, all with unlimited users and workflows. The Community Edition is free to self-host from GitHub.
The product is a builder for engineers and technical operations staff more than a workspace for business users, so it suits teams that want triggers, branching and code next to agent steps.
What n8n does well
- Cloud data hosted in Frankfurt, per the pricing page
- Priced per execution, with unlimited users, workflows and steps
- Free self-hosted Community Edition
- Full legal set: Terms, AI Terms, DPA, security page and Trust Center
Where n8n falls short
- A workflow builder first; less suited to business users who want a chat-style agent library
- Business plan at €667 a month is self-hosted only
- Model providers used inside workflows are the customer's choice and may sit outside the EU
- Licence terms were not verified for this page
Standout feature. It is the only platform here that prices by execution and also publishes that cloud data sits in Frankfurt.
Paris, France
Free / Pro $14.99 per month / Team $24.99 per user per month / Enterprise on request, with custom agents listed under Enterprise / API billed by usage
Free tier
Best for: Developers who want an agents API from a European model provider
Mistral AI documents an Agents and Conversations API for building agents with persistent state across conversations, handoffs between agents, and built-in connectors for code execution, web search, image generation and a document library. The Studio documentation groups conversations, agents, connectors, workflows, search and observability.
The terms name Mistral AI as a French limited joint-stock corporation incorporated in Paris under number 952 418 325, with its registered office at 15 rue des Halles, 75001 Paris.
Its Help Center states that by default data is hosted in the European Union, that a US API endpoint can be used instead, and that data can be temporarily transferred outside the EU to sub-processors listed in its Trust Center. Governing law varies by customer region, with California law for customers in the Americas.
The pricing page lists Free at $0, Pro at $14.99 a month, Team at $24.99 per user per month and Enterprise on request, where it names custom models, custom agents and custom workflows. The API is billed by usage.
What Mistral AI does well
- French company with registration number and address in its terms
- EU hosting by default per the Help Center
- Agents API with handoffs, persistent state and built-in connectors
- Prices in US dollars published for the assistant plans
Where Mistral AI falls short
- Custom agents appear only under the Enterprise plan on the pricing page
- Temporary transfers outside the EU to sub-processors are allowed
- Prices are shown in US dollars
- Mostly its own models, so no choice of provider
Standout feature. It is the only platform here that builds the agents on models it trains itself, in the same company that hosts the data.
Heidelberg, Germany
Not published, contact sales
Contact sales
Best for: Public bodies and enterprises that need an on-premises or air-gapped agent stack
Aleph Alpha sells a platform described as a sovereign foundation to build, scale and optimise AI solutions, available on-premises, air-gapped or as a sovereign SaaS. Its platform page lists Document AI for parsing and retrieval, with answers that quote source passages and abstain when the evidence is insufficient, and an Agent Orchestration and Runtime for composing agents from reusable skills with approval gates and logged actions.
The imprint names Aleph Alpha GmbH at Speyerer Straße 14, 69115 Heidelberg, registered at the Amtsgericht Mannheim under HRB 732804 with VAT ID DE322477849 and Ilhan Scheer as the representative. Further offices are listed in Berlin, Bayreuth and Munich.
No price is published and the site routes every enquiry to sales. No hosting region is stated for the SaaS option; the on-premises and air-gapped options leave hosting to the customer.
What Aleph Alpha does well
- On-premises, air-gapped or sovereign SaaS deployment
- Approval gates and logged actions in the agent runtime
- Document AI that abstains rather than guesses
- German GmbH with a full imprint
Where Aleph Alpha falls short
- No published price
- No hosting region named for the SaaS option
- Sold as a project through sales, with no self-serve entry
- No certification named on the pages read
Standout feature. It is the only platform here that offers an air-gapped deployment as a standard option.
Amsterdam, Netherlands
Open Access free / Personal AI Assistant €20 per month billed annually / Team and Enterprise on request
Free trial on request for the personal plan
Best for: Research-heavy organisations that want agents on top of enterprise search
Zeta Alpha describes itself as an enterprise AI search and agentic RAG platform.
Its pricing page lists an Open Access plan that searches open AI research for free, a Personal AI Assistant at €20 a month billed annually, a Team plan for private and team information from tools such as Trello, Slack, Jira, Confluence, Google Drive and SharePoint, and an Enterprise plan with custom AI agents on private data and a secure private tenant with an option to host on premises.
The terms of service and privacy policy name Zeta Alpha Vector B.V., organised under the laws of the Netherlands, with its registered office at Science Park 608, 1098 XH Amsterdam. The privacy policy covers the website and the AI Research Navigator service.
The pages read give no hosting region or cloud provider, and the Team and Enterprise plans are quote-only. The Enterprise plan lists a 99.9% availability guarantee.
What Zeta Alpha does well
- Dutch B.V. named in both terms and privacy policy
- Custom agents on private data on the Enterprise plan
- Private tenant with an on-premises option
- Federated search across common team tools
Where Zeta Alpha falls short
- No hosting region or provider named on the pages read
- Team and Enterprise pricing is quote-only
- Search is the core; agent building is an Enterprise feature
- No security certification named on the pages read
Standout feature. It is the only platform here that starts from enterprise search and a free open-access research product before the agent layer.
Berlin, Germany
Studio free / Enterprise on request
Free Studio: 1 workspace, 1 user, 100 pipeline hours, 50 files
Best for: Developers building retrieval and agent pipelines with a Berlin vendor
deepset is the Berlin company behind the Haystack Enterprise Platform, which its pricing page describes as a unified platform to develop and deploy LLM applications and agents. The plans list a Pipeline Builder, Prompt Explorer, observability, groundedness checks, API access and, on Enterprise, role-based access control, SSO and custom components.
The imprint names deepset GmbH at Zinnowitzer Str. 1, 10115 Berlin, registered at the Amtsgericht Berlin-Charlottenburg under HRB 197429 B with VAT ID DE319210575 and Milos Rusic and Malte Pietsch as managing directors. The terms add that the contracting party is deepset Inc. in New York for customers located in the United States and deepset GmbH for customers elsewhere.
Studio is free with one workspace, one user, 100 pipeline hours, 50 files and two development pipelines on cloud deployment. Enterprise is custom, with cloud or custom deployment and flexible deployment options on request. No hosting region is named on the pages read.
What deepset does well
- German GmbH with a complete imprint
- Free Studio plan with defined limits
- Cloud or custom deployment on Enterprise
- Observability and groundedness checks listed in the feature table
Where deepset falls short
- US customers contract with deepset Inc. in New York
- No hosting region named on the pages read
- Enterprise pricing is quote-only
- Free plan is limited to one user
Standout feature. It is the only platform here whose terms spell out a different contracting entity for US customers.
Berlin, Germany
Self-managed development free / self-managed production and SaaS on request
30-day SaaS trial, any usage during it never billed
Best for: Organisations that already run business processes and want agents inside them
Camunda describes itself on its homepage as the open platform for agentic orchestration, orchestrating AI agents, people and systems across end-to-end business processes with governance and control. It is a process orchestration platform first, and agents are one of the things it coordinates.
The imprint names Camunda Services GmbH at Zossener Straße 55-58, 10961 Berlin, registered at the Amtsgericht Charlottenburg under HRB 113230 B with VAT ID DE259411084 and Jakob Freund and Clemens Morgenroth as managing directors.
The pricing page makes development on your own machine free with no licence, and requires an Enterprise licence and a sales conversation for self-managed production and for SaaS. SaaS has a 30-day free trial in which usage is never billed. It states that the SaaS Console and Modeler are hosted in the EU and that clusters can be created in other regions.
What Camunda does well
- German GmbH with a complete imprint
- SaaS console and modeler hosted in the EU, per the pricing page
- Free self-managed development with no licence
- 30-day SaaS trial that is never billed
Where Camunda falls short
- No published production price
- Clusters can be created outside the EU, so the region needs checking
- A process engine first, so more work than a workspace for a simple agent
- Requires engineering effort to model processes
Standout feature. It is the only platform here that frames agents as one participant among people and systems in a modelled business process.
Zurich, Switzerland
Best for: Regulated enterprises looking for a Swiss enterprise AI vendor with an on-premises option
Squirro presents an enterprise AI platform whose navigation lists an Agentic AI engine, Knowledge Graphs, AI Guardrails, a Privacy Layer, an AI Agent Catalog, a Security and Trust Center and On-Premises Enterprise AI. Its banner invites visitors to explore the Squirro AI Agent Catalog.
The privacy policy names Squirro AG, with its postal address at Mühlebachstrasse 164, 8008 Zurich, Switzerland, as the legal entity. Switzerland is in Europe but outside the EU and the EEA, so EU law does not apply to it directly.
The pages read publish no price, no hosting region and no certification, and the trust center page could not be read. The entry in this list rests on the entity and the product navigation only.
What Squirro does well
- Swiss AG with a postal address in its privacy policy
- Agentic AI engine and an agent catalog in its navigation
- On-premises enterprise AI listed as an offering
Where Squirro falls short
- No price published on the pages read
- No hosting region or certification verified
- Switzerland is outside the EU and EEA
- The privacy policy read carries a 2018 revision date
Standout feature. It is the only platform here that lists a ready-made AI agent catalog next to an on-premises option.
Dublin, Ireland
Free edition with 3 live workflows, unlimited users and a one-time $50 AI allowance / paid editions on request
Free edition
Best for: Security and IT teams adding agents to workflows they already govern
Tines is a workflow platform that positions itself as letting every team build AI-native agents, apps and automations while IT and security keep visibility, governance and control. Its navigation lists Tines 3B, Agents, Workbench, Cases and Monitoring.
The terms say the contracting entity depends on the customer's place of business: a Delaware corporation for customers in North America and an Irish company for customers outside it, with Irish law and Irish courts for the Irish entity. The data processing agreement names Tines Security Services Limited at The Academy, 42 Pearse Street, Dublin D02 YX88, Ireland. The terms use more than one name for these entities, which is worth checking against the order form.
The pricing page lists a free edition with unlimited users, spaces and connectors, 3 live workflows and a one-time $50 AI allowance, and paid editions with unlimited workflows and the option to bring your own AI model. Paid prices are not shown. The security page says Tines is a cloud service it hosts and a product customers can host in their own data centre, and names no cloud region.
What Tines does well
- Irish company named in the DPA with a Dublin address
- Free edition with unlimited users
- Bring-your-own AI model on paid editions
- Cloud or deployment in your own data centre
Where Tines falls short
- A Delaware company contracts with customers in North America
- No hosting region named on the pages read
- Paid prices are not published
- Entity names in the terms are inconsistent
Standout feature. It is the only platform here whose free edition is capped by the number of live workflows rather than by users.
Spain
Not published: fees as stated in the order form, admin console or plan
Meeting booked through the site; no self-serve trial found
Best for: Companies that want agents to prepare operational decisions with a recorded approval
Hellomatik is a Spanish platform, called HM-2, built on a company blueprint: a written description of the company's data, procedures, formulas and permitted actions. AI agents prepare each decision on that blueprint, and an approver accepts, amends or rejects it. The site lists use cases such as supplier invoice posting, pre-call account review and email order processing, with every run kept in an operations log and every approval in an approval log.
The terms of service name HELLOMATIK, S.L. (CIF B22803126) as the contracting entity, under the laws of Spain with the courts of Madrid, and the privacy policy names the Spanish AEPD as supervisory authority.
The site states that data is encrypted and hosted in the European Union, with servers in Germany. Its language-model providers are contracted as sub-processors under Article 28 of the GDPR, none of them trains models on company data, and the list with locations is available on request.
No price is published: fees follow the order form, admin console or plan, and usage-based charges such as minutes and messages are billed on actual consumption. The entry point is a booked meeting in which the company names its systems and the first decision it wants prepared.
What Hellomatik does well
- Spanish S.L. with a CIF in its terms and Spanish law governing
- Servers stated to be in Germany
- An approval log for every decision an agent prepares
- States that no language-model provider trains on company data
- Built around company systems such as ERP, email and warehouse
Where Hellomatik falls short
- No published price or self-serve trial
- Sub-processor locations for model calls are available only on request
- Use cases are built per company, so time to value depends on the first project
- No security certification named on the pages read
Standout feature. It is the only platform here where the approver and the approval log are the core of the product rather than an optional gate.
