Weloop
Weloop is a French AI agent that sits inside a company's internal applications, Teams and Slack. It answers staff questions from their own knowledge base with citations, qualifies requests, opens the right tickets and tracks satisfaction.
What is Weloop?
Weloop is an AI agent embedded in the internal applications a company already runs, marketed as an AI Product Manager or a delegated product manager. Employees write in plain language inside a widget that lives in web applications built with React, Vue.js or Angular, in Microsoft Teams or in Slack; a single script installs it. The agent already knows which page the user is on, so it confirms the context instead of asking for it.
What happens next follows four movements, with a fifth optional step. It engages, asking the follow-up questions that would otherwise cost the team a round trip. It resolves, searching the customer's knowledge base and answering with the source article cited; simple cases close end to end without anyone stepping in, and duplicates are recognised and merged once the reporter agrees. When a bug is hard to describe, the agent offers to record the screen: consent is explicit, passwords and sensitive fields are masked automatically, and the analysis pre-fills the form. It then decides and routes, gathering the fields your business rules require and opening the ticket in Jira, ServiceNow, Slack, Teams, by e-mail or in the built-in kanban, with mapped fields and two-way reply syncing. Finally it follows up, measuring first-response and resolution SLAs in working hours only, and closing with an NPS or CSAT question.
Around the agent sit a knowledge base fed by AI import of existing PDF, Word, PowerPoint and Markdown files up to 40 MB with every draft reviewed by a person before publication, business rules described in plain French or English rather than code, an announcements and changelog module, survey tooling, and a monitoring view of what the agent did. The agent also proposes changes to its own configuration, each scored for confidence and risk, approved by a human, versioned with a before-and-after diff and reversible.
The publisher builds its pitch on sovereignty: data stored in France on Scaleway, Mistral AI models, DeepL for translation, and three deployment modes including full self-hosting on Kubernetes or AWS. Performance figures such as 92% autonomy or 11 hours saved per week are the vendor's own claims, not audited measurements.
What it does
- Answers employee questions instantly from the company's own knowledge base, citing the article it used
- Detects duplicate reports and merges them once the user agrees, turning the number of affected people into a priority signal
- Collects the fields required by business rules written in plain language, then opens a ticket in Jira, ServiceNow, Slack, Teams, e-mail or its built-in kanban
- Hands over to a human in one message, with the handover request detected in any language
- Records the user's screen with explicit consent and automatic masking, then uses the recording to pre-fill the report
- Runs NPS, CSAT and custom surveys in the widget, by e-mail or embedded in the customer's own pages
- Proposes its own configuration improvements with a confidence and risk score, each approved by a human and reversible
When to use Weloop / When not to
A quick filter to help you decide if Weloop is the right fit.
When to use Weloop
- Product managers and product owners who currently spend their week triaging support tickets for an internal application instead of building it
- IT and digital-workplace teams running an internal service desk that keeps answering the same questions
- CIOs and DPOs in regulated European organisations (banking, energy, industry, public sector) who need data to stay in France or inside their own infrastructure
- Business-application owners in HR, finance or operations who want employees helped inside the tool they already use, with no new interface to learn
- Customer-experience and user-research teams who want NPS, CSAT and surveys tied to requests that actually get resolved
When not to use Weloop
- Individuals and small teams: the entry tier is 590 EUR per month for one internal application, with no free plan and no advertised trial
- Companies wanting to equip a single low-traffic application, since the price is charged per internal application rather than per company
- Public-facing customer support: Weloop is aimed at a company's own employees, not at external consumers
- Organisations needing an interface in a language other than French or English
- Buyers who need a formal security certification on file, since Weloop claims an annual independent audit but no ISO 27001, SOC 2, HDS or SecNumCloud credential
How to use Weloop
A typical end-to-end flow, from setup to results.
- Book the 30-minute demo, which is the only entry point: there is no self-service sign-up
- Drop a single script tag loading js.weloop.ai into your internal application, or install the @weloopai/widget npm package and call init with your project ID
- Import your existing PDF, Word, PowerPoint and Markdown documents so the AI drafts knowledge base articles, then review each draft before publishing it
- Describe your business rules in plain language: which information to collect and where each request should go, then refine the configuration the assistant produces
- Connect the destinations you use, among Jira, ServiceNow, Slack, Microsoft Teams, e-mail and the built-in kanban, and map the fields
- Wire up authentication, choosing between SSO with SAML 2.0 or OIDC, e-mail magic links, or a server-signed JWT passed to the widget for silent identification
- Set your SLA targets in working hours and configure the NPS, CSAT and survey triggers
- Let employees write to the widget and watch the monitoring view for deflection, routing and human handovers
- Review the improvements the agent proposes, approve or reject each one, and roll any change back from the history if needed
- For a self-hosted deployment, provision a Kubernetes cluster with PostgreSQL, a Redis-compatible cache and S3-compatible storage, then install the Helm chart with the licence key Weloop issues
Pros & Cons
Pros
- Sovereignty is engineered rather than asserted: French hosting on Scaleway, Mistral AI models, DeepL translation, and a genuinely documented self-hosting path with Helm and Terraform
- Public technical documentation is unusually thorough for a product of this size, including a detailed security page listing log retention, RTO and RPO targets
- Pricing is published and readable, with two figures on the page and only the largest tier left to quotation
- Billing per internal application rather than per ticket or per administrator seat makes the cost predictable as usage grows
- Governance safeguards are real: every configuration change, human or AI-proposed, is logged, attributed and reversible, and knowledge base drafts are reviewed before publication
- SLAs measured in working hours mean nights and weekends do not count against the team's clock
- A dedicated Customer Success Manager is included in every tier, and an external DPO is registered with the French regulator
Cons
- No free plan and no advertised trial, with an entry point of 590 EUR per month
- The price applies per internal application, so equipping several applications multiplies the bill in a way the headline figure does not show
- No legal notice page exists: the mentions-legales URL returns a 404, and the company identity appears only in the privacy policy, the terms and the structured data
- The site contradicts itself on availability, claiming 99.5% uptime on the marketing pages and in the documentation but 99.99% in the terms of use
- The privacy policy is a generic template, titled after a different domain, that states it does not list the categories of data collected and gives no retention period
- No subprocessor list is published: the privacy policy refers you to the vendor and the terms refer to a contractual annexe
- Nothing is said about whether customer data trains AI models, and no opt-out is documented, while the terms grant an irrevocable right to reuse anonymised user contributions
Pricing & Plans
There is no free plan and no free trial is advertised. The cheapest paid entry point is the Demarrage tier at 590 EUR per month for a project with fewer than 50 users. The essential caveat is the unit: Weloop states that the price is charged per internal application equipped, not per company, per ticket or per administrator seat, so a second application means a second subscription. The Croissance tier costs 1,490 EUR per month for projects up to 500 users, and the Entreprise tier is quoted on request above that. Every tier includes the full agent, sovereign French hosting, GDPR compliance and a dedicated Customer Success Manager.
- 590 EUR per month - project with fewer than 50 users - dedicated Customer Success Manager included
- 1
- 490 EUR per month - project of up to 500 users - flagged as the recommended tier - dedicated Customer Success Manager included
- price on request - project with more than 500 users - dedicated Customer Success Manager included
Data, GDPR & hosting
A consolidated view of how Weloop handles your data.
GDPR overview
Implementation is concrete and above average for a company this size. Weloop is French, the contract is governed by French law, and the terms place the customer as controller and Weloop as processor. Named commitments: a DPA available on request at gdpr@weloop.ai; an external Data Protection Officer, BDO Strategie et Performance, registered with the French CNIL under reference DPO-118888; rights exercised at privacy@weloop.io or by post to Frontbox in Saint-Mande, answered within one month; no transfer of personal data outside the EEA without the customer's prior authorisation and three months' notice; an annual on-site audit right for the customer; and a processor register available on request. Screen recording runs on explicit consent with automatic masking. The weak point is the privacy policy itself, a generic template that lists neither the data categories collected nor any retention period.
Who owns the data?
The customer keeps ownership. Under the terms of use the customer is the data controller and Weloop acts only as its processor, and the customer is stated to be solely responsible for the Content exchanged by its users. Weloop receives a hosting, reproduction and display licence limited to the subscription term, and it also takes a broader right: an irrevocable, unpaid licence to incorporate user suggestions, improvement requests and comments into its own Services once anonymised. On termination, and at the customer's written request, Weloop must either destroy all personal data or return it in a readable, machine-usable format, deleting remaining copies unless the law requires otherwise.
Reuse rights
Customers cannot freely reuse whatever they like: the licence Weloop grants is a personal, non-transferable right to use the Services and reproduce the Weloop script for the duration of the subscription, and nothing beyond that is conceded. In the other direction, Weloop undertakes not to use personal data for any purpose outside the contract, and explicitly not for prospecting or marketing, and not to sell, rent or pass it to a third party. Billing and reporting telemetry is stated to carry no personally identifiable information. The exception to watch is the anonymisation clause: once user contributions are anonymised, they leave the confidentiality perimeter and Weloop may fold them into the product without compensation. Nothing on the site states whether customer data is or is not used to train AI models, and no opt-out mechanism is documented.
Data retention & training
Hosting summary
Weloop stores customer data in France, on Scaleway infrastructure in the Paris region, and describes the surrounding stack, including the CDN and peripheral services, as fully European. The AI models come from Mistral AI and translation runs on DeepL, with the vendor stating that no data reaches United States providers. Three deployment modes exist: the Weloop cloud operated by the vendor in France, the same platform served under the customer's own domain, and full self-hosting on the customer's infrastructure, where the data never leaves their information system and the customer supplies PostgreSQL, a Redis-compatible cache and S3-compatible object storage. Traffic is encrypted with TLS 1.2 or higher and plain HTTP is refused. Production data is backed up daily with a 30-day retention and quarterly restore tests, against a 4-hour recovery time objective and a 24-hour recovery point objective. The terms of use forbid any transfer of personal data outside the European Economic Area without the customer's prior authorisation and at least three months' notice.
Things to keep in mind
Risks and trade-offs to weigh before adopting Weloop.
- The advertised price is per internal application: count the applications you actually intend to equip before comparing Weloop with a per-seat competitor
- The site contradicts itself on uptime, promising 99.5% on the marketing pages and in the documentation but 99.99% in the terms of use, so ask which figure is contractually binding
- The terms grant Weloop an irrevocable, unpaid right to fold anonymised user contributions into its product, and anonymised content sits outside the confidentiality clause
- Nothing is published about whether customer data trains AI models, and no opt-out exists, which is a question a data protection officer will ask first
- No subprocessor list is published and no legal notice page exists, so the vendor's supply chain and legal identity both have to be requested or inferred
- Scaleway, Mistral AI and DeepL are Weloop's suppliers: their own guarantees and certifications belong to them and should not be read as certifications held by Weloop
- An agent that closes 92% of requests without a human also removes the informal contact through which teams used to sense what was going wrong, so keep reading the monitoring view rather than trusting the deflection rate alone
Setup & Integrations
Technical difficulty
Low in the cloud, high if self-hosted. A cloud rollout needs one script tag pointing at Weloop's CDN, or the npm package and a single init call; business rules are then written in plain language rather than code, and a dedicated Customer Success Manager is included. A developer is only required for silent identification, which means signing short-lived JWTs server-side, and for SSO and SCIM setup with your IT department. Self-hosting is a different project altogether: a Kubernetes cluster, PostgreSQL, a Redis-compatible cache, S3-compatible storage, ingress and TLS, SMTP and a licence key.
Deployment
Integrations
Supported languages
Behind Weloop
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What exactly does Weloop do?
How much does it cost, and is there a free plan?
Where is the data hosted?
Can it run on our own infrastructure?
Which tools does it integrate with?
Is there an API for developers?
Is Weloop GDPR compliant, and is a DPA available?
Does Weloop use our data to train AI models?
What availability does Weloop commit to?
Who publishes Weloop?
Should you pick Weloop?
Weloop is a focused, credible tool for a specific pain: an internal application whose product or project team has quietly become its help desk. The mechanics are coherent, from context-aware answering and duplicate merging through to routing into Jira or ServiceNow and closing the loop with an NPS question, and the governance details suggest real product maturity rather than a demo, with every configuration change attributed and reversible and every AI-written article reviewed by a person.
Its sovereignty claim is the strongest part and it survives inspection layer by layer: hosting on Scaleway in France, Mistral AI models, DeepL translation, and a self-hosting path documented well enough to be evaluated by an infrastructure team. For a regulated European buyer, that is a genuine differentiator rather than a slogan.
What lets the product down is the paperwork around it. There is no legal notice page at all, the privacy policy is a generic template carrying a different domain name and no retention periods, no subprocessor list is published, and the uptime commitment differs between the terms of use and every other page. None of this makes the product worse, but all of it makes due diligence harder than it needs to be, and a DPO will notice.
Two practical points before you compare quotes. The price is per internal application, so model the real number of applications you intend to equip rather than the headline figure. And every performance number on the site, from the 92% autonomy rate to the claimed sevenfold to twelvefold return, is the vendor's own; ask to see them measured on a pilot before treating them as a business case.
- Choosing a selection results in a full page refresh.
- Opens in a new window.