Frequently asked questions
What VasperaMesh is, how the tamper-evident ledger works, and how it compares.
What is VasperaMesh?
VasperaMesh is the deterministic control plane for regulated AI decisions. You route your AI agents' model calls through the VasperaMesh gateway, and every decision becomes one immutable row in a hash-chained audit ledger that is externally anchored to Bitcoin and verifiable offline — without trusting the vendor. Policies are authored in plain English and enforced by deterministic gates that return pass, hold, or escalate.
How do I connect my agents? Do I have to rewrite my code?
No rewrite. If your agents already emit OpenTelemetry (the GenAI semantic conventions — invoke_agent, chat, execute_tool spans, which frameworks like the Vercel AI SDK, LangChain, and CrewAI produce), you point your existing OTLP exporter at VasperaMesh — the same one line you would send to Datadog or Honeycomb. That captures the actual decisions (tool calls and structured outputs), not just generations, and gates them. For callers not on OpenTelemetry, a shadow proxy gives you the same audit trail with a one base-URL change.
How does VasperaMesh help with the new Fannie Mae and Freddie Mac AI rules?
Fannie Mae Lender Letter LL-2026-04 (effective August 6, 2026) requires seller/servicers using AI/ML in origination or servicing to maintain documented, actively-maintained governance and to measure and manage AI/ML risk; Freddie Mac's framework (Bulletin 2025-16) is already in effect. VasperaMesh gives you a live, auditable governance record — every AI decision recorded, versioned against the policy in force, and reproducible — plus one-click evidence packs mapped to those requirements. Because the record is cryptographic and offline-verifiable, it exceeds a documentation-only program, and it produces the specific, reproducible adverse-action reasons ECOA / Reg B require without relying on the model to explain itself.
How does the tamper-evident audit ledger work?
Each agent action — a model generation, a gate verdict, a policy change, a human override — is written as one append-only row. Rows are canonicalized with RFC 8785 (JCS), domain-separated, and hashed with SHA-256 into a chain, so editing any earlier row breaks every hash after it. Chain heads are timestamped with OpenTimestamps and confirmed on the Bitcoin blockchain, and database triggers reject updates and deletes (write-once, read-many).
How can I verify a decision without trusting VasperaMesh?
A standalone verifier re-derives every row hash from its canonical fields and cross-checks the Bitcoin anchors, importing nothing from VasperaMesh. Anyone can verify a chain at https://www.vasperamesh.com/verify with no login, or download the verifier script and run it themselves. The published ledger specification is at https://www.vasperamesh.com/spec.
How is VasperaMesh different from observability tools like LangSmith, Langfuse, or Arize?
Observability tools trace agent runs but store mutable, vendor-trusted logs — they cannot cryptographically attest to a decision without breaking their own model. VasperaMesh records each decision at the moment the action is proposed into a hash-chained, externally anchored ledger that a regulator can verify offline. It is a runtime enforcement point that also emits an un-fakeable record, not just a trace viewer.
How is it different from AI governance / GRC platforms like Credo AI or Monitaur?
GRC platforms document governance, but their evidence is self-attested and most lack a runtime enforcement point. VasperaMesh pairs plain-English policy authoring with a runtime gate and backs every control with the cryptographic ledger, so compliance evidence is provable rather than asserted.
Which regulations and frameworks does VasperaMesh help with?
VasperaMesh generates auditor-ready evidence packs mapped to the EU AI Act (Articles 12, 14, and 19), ISO 42001, and NIST AI RMF, and supports the tamper-evident, attributable, retained audit trails required by regimes such as SR 26-2 / ECOA for lending, FINRA/SEC record-keeping, and HIPAA. Every control in a pack is backed by the hash chain, not self-attestation.
How much does VasperaMesh cost?
There is a Free plan with 5,000 ledger rows per month and one policy; a Pro plan at $99 per month with 100,000 ledger rows per month, unlimited policies, and backtests; and a custom Enterprise plan with self-hosted / VPC deployment, SSO, and WORM anchoring. See https://www.vasperamesh.com/pricing.
Do you store our prompts and model outputs?
By default VasperaMesh stores only cryptographic digests of prompts and responses, not their contents. Content retention is opt-in per organization; when enabled, content is stored with AES-256-GCM envelope encryption and can be pruned on a schedule for data-retention and erasure requirements, while the tamper-evident row, digest, and Bitcoin anchor remain permanent.
How do I get started?
Sign up and create an organization, author a policy in plain English and have a second person approve it (separation of duties is enforced), mint an API key, and point your agent's model base URL at the VasperaMesh gateway. Decisions begin landing in the ledger immediately, and you can verify any of them at /verify with no login.