algovigilance.com

What it is

algovigilance positions itself as an “auditable algorithmic vigilance” foundation layer, rather than a standard chatbot or content generation tool. Its core argument is that every algorithmic decision should come with a receipt: what decision was made, why it was made, whether the system had sufficient confidence, and whether it should have refused to answer. Its target audience is clearly tilted toward boards, auditors, regulators, and AI/software teams in regulated industries.

Core capabilities

Based on the information on the site, the product centers on a decision kernel and an audit-evidence layer. Every published decision is signed with EdDSA; if the logic is changed, the signature becomes invalid, making it easier to prove consistency across development, production, and regulatory review scenarios. It also emphasizes a plain-English trail, so non-technical stakeholders can read the decision records. For LLM use cases, the system structures the audit process around whether the response followed the kernel verdict; if the LLM ignores the verdict, the kernel catches it instead of relying on manual review. Another key design principle is that it “refuses to guess”: when input falls outside the declared domain, it refuses directly and explains why.

Pricing and commercial status

No plans, pricing, free quota, or trial policy are currently disclosed. The site indicates that the product is still in live-testing, with sales expected to open in Q3 2026. Its business model is described as follows: the decision substrate is open and self-hostable, while the commercial product consists of hosted edges, including SLAs, capability suites, and certification. The terms of service are also still a placeholder version, and the formal commercial agreement has not yet been released.

Pros and cons

Its strengths are its very clear positioning and the fact that it addresses the audit questions that matter when deploying regulated AI: who made the decision, why it was made, whether it was tampered with, and when the system should refuse. Signatures, versioning, queryability, and readable trails are valuable for compliance communication. The drawbacks are also obvious: there is no detail on model capabilities, performance metrics, deployment documentation, real customer cases, or privacy policy specifics. Chinese-language support is not mentioned, and the site emphasizes English audit trails.

Who it is for and access from China

It is best suited to teams in finance, healthcare, government, enterprise risk control, and other fields that need traceable algorithmic decision records—especially organizations that already have AI systems but lack a verifiable governance layer. Access from mainland China, payment methods, and local compliance adaptation are all undisclosed and should currently be treated as “unknown.” If you need a mature alternative in the short term, you may want to consider a combination of existing MLOps, model governance, LLM guardrails, audit logging, and compliance-oriented model management tools, though specific alternatives should be chosen according to the regulatory requirements of each industry.