Langfuse
Vannus records that this vendor routes across several model providers — not captured by one. Every finding below is quoted to the vendor’s own document, or marked not disclosed where the vendor publishes nothing.
LLM observability and analytics platform — session-level tracking of intent, coherence, tool-call accuracy, and context retention, catching the coherence trap before agents drive off logical cliffs
The Playground and LLM-as-a-Judge evaluations can be used by any framework that supports the OpenAI API schema such as Groq, OpenRouter, Vercel AI Gateway, LiteLLM, Hugging Face, and more.
Training and retention posture varies by plan. What we publish above describes the vendor’s default plan; enterprise, team and API agreements frequently differ, often materially, and a contract can override the published default entirely. Check your own plan and contract before relying on this row.
This is a separate question from the grade above. The grade measures resilience — whether the tool endures and whether you could leave it. This describes who controls the vendor. A tool can score modestly on one and strongly on the other, and many do.
On U.S. CLOUD Act reach specifically: the statute reaches a provider subject to U.S. jurisdiction over data in its possession, custody or control. Corporate control is a strong indicator of that and it is what we can evidence from published documents — but it is not the whole test. A company founded outside the U.S. can still contract through a U.S. entity or run substantial U.S. operations. Treat this as a starting point for your own review, not a legal determination, and take advice on anything that matters.
The vendor's published or catalog-recorded posture — the concrete facts this grade is built from. A full audit verifies each against the vendor's current documentation.
Langfuse publishes a list of the other companies it uses to process customer data. It names 8 of them, each shown below with the location the vendor lists it under, in the vendor’s own words.
Taken together those entries name at least United States of America. That is what our place list could match in the vendor’s own words above, so treat it as a floor rather than the whole of it — the entries themselves are the record.
These are other companies, not Langfuse. Where a sub-processor is listed as operating is a fact about that company. It is not a statement about where Langfuse keeps your data, which Vannus publishes separately and only from a document in which the vendor says so.
Read from clickhouse.com on 2026-09-16. Every entry above is a verbatim span of that page.
This is the vendor’s own disclosure, reproduced. Vannus has not audited what any of these companies do with your data, and a list can change without notice. Treat it as a starting point for your own review, not a legal determination, and take advice on anything that matters.
Vannus publishes a nine-dimension trust framework — data sovereignty, training privacy, compliance posture, operational resilience, exit portability, and more. The heaviest criterion asks whether the tool builds its own AI or resells someone else's model; where the vendor discloses it, the grade cites the vendor's own documentation. No paid placements — scoring is walled off from affiliate revenue. See the methodology →