Building with
A post-hoc log can be disputed. A pre-execution attestation can't. AqtaCore produces a signed verdict before the model responds, every inference call. One drop-in API, across major LLM providers.
Every model invocation is intercepted, checked against your policy, and cleared or blocked. Bad requests never reach the provider. Budget and rate limits hold. A signed record is created either way.
Pre-flight · Tool calls · Budget limits · Drop-in API
Every cleared action gets an Ed25519-signed verdict at inference: content hash, policy version, timestamp. Receipts hash-chained at audit export. Hand it to a regulator. Hand it to a court.
Signed at inference · Chained at export · Hand to a regulator
Every agent run is a tracked session: steps, cost, status, timestamps. Cut it at any point and get a receipt backed by the Ed25519-signed traces underneath.
Session kill · Full audit · Backed by signed traces
Articles 9, 11, and 13 require risk management records, technical documentation, and transparency logs. AqtaCore generates all three automatically, exportable on demand.
Article 9 · Article 11 · Article 13 · Export on demand
AqtaCore signs AI decisions. Bounds Pro signs document handling. Two surfaces, one evidence layer.
The runtime attestation gateway for agentic AI. One drop-in API, kill any session, enforce policy at inference, cryptographic audit trail across major LLM providers.
On-device AI PDF redaction with tamper-evident audit trails. Five AI models run entirely in the browser via WebAssembly. Nothing touches a server.
Aqta Labs
Other products we've shipped along the way.
Adaptive test prep for GMAT, GRE, SAT, LSAT, IELTS.
Hands-free voice browsing, online or offline.
Zoonotic spillover risk mapping. 53-day median lead ahead of WHO notification in backtests.
AI companion that shows its work. Your receipts, your key.
AqtaCore
Every decision leaves an audit trail: hash, policy version, timestamp.
Generated at inference. Not reconstructed after.
