Product
You should be able to explain every reply your company sends
logs every decision, every data source, and every policy check. When a customer asks why they received a particular answer, you can show them exactly how it was produced.
Ifyoucannotexplain
whyareplywassent,
youshouldnothave
sentit.
Audit trail
Every step recorded. Every source cited.
For every conversation, the platform records which systems were queried, what data was returned, which knowledge base entries were consulted, which policies were applied, and how the validation pipeline assessed the reply. The full trace is available in the dashboard.
System queries
Every MCP tool call is logged with the request, the response, and the time it took. You can see exactly which data from your ERP, CRM, or order system informed the reply.
Policy application
Which company policies were evaluated, whether the reply passed or failed each check, and if it was rewritten -- what feedback was given and how the next draft differed.
Validation pipeline
Every reply is checked before it reaches the customer
A separate validator fact-checks every claim in the reply against the source data retrieved from your systems, verifies compliance with your policies, and enforces your brand voice. Replies that fail verification are rewritten with specific feedback. When verification cannot be completed after three attempts, a team member takes over and completes the conversation. The platform delivers only what it can verify; the team handles what requires their judgement.
Autonomousdoesnotmeanunsupervised.
Itmeansconsistentlyaccountable.
Data residency
EU-hosted. All processing, all storage, all inference.
All infrastructure and AI inference runs in the European Union. Platform hosted in Germany; head office in Copenhagen. Customer data, conversation history, knowledge base content, and AI inference all stay within the Union. There is no option to move processing outside the EU because there is no infrastructure outside the EU. Enterprise deployments can run LLM inference on your own infrastructure via the private-inference option, in which case conversation data never enters any shared inference environment at all.
Account isolation
Your data is yours. Completely separated.
Every database query, every storage path, every cache key, and every workflow execution is scoped to your account. There is no shared data layer between customers. One account's data cannot be accessed by another account at any level of the stack.
Privacy by design
No cookies. No tracking. No third-party scripts.
This website does not use cookies or analytics. The platform does not share data with advertising networks. Fonts are self-hosted. There are no third-party tracking scripts on any page. Your visitors' privacy is respected by default, not by opt-out.
Human oversight
Your team stays in control
Approval gates pause execution for high-value actions. Takeover rules route conversations to your team when specific conditions are met. The learning loop only updates knowledge and policies after human review. The platform handles the volume it can verify. Your team handles the judgement.
Compliance
Built to meet European regulation by construction
The five-layer safety model is designed to satisfy the record-keeping obligation of the EU Artificial Intelligence Act (Regulation (EU) 2024/1689, Article 12) and the human-oversight obligation (Article 14) by construction. Controllers obligations under the General Data Protection Regulation (Regulation (EU) 2016/679) -- the accountability principle (Article 5(2)) and records of processing activities (Article 30) -- are met architecturally, not by retrofitted policy. A Data Processing Agreement compliant with Article 28 is available for every customer.
See the full audit trail in action
Connect your systems and trace every decision from query to reply