Aurra

Aurra is audit-grade memory for AI agents. It is built for cases where an agent’s recall has to be trusted, traced, or proven, and it gives each answer a receipt showing what the system knew, when it knew it, and why it responded as it did.

Its main features are centered on verifiability and time-aware recall. Aurra returns a proof tree with every answer, including a structured record of the exact memories used, a tamper-evident hash, and what was superseded. It is bi-temporal, tracking both when a memory was captured and when it was true, and it can query the state of memory at a point in the past. The product also keeps source citations and a full audit trail, with an audit endpoint that includes the extraction model, prompt version, and event history. The page says every memory and every answer is recorded in a tamper-evident way.

Aurra is described for teams that need memory that is trusted and traceable, from regulated enterprises to solo builders. The examples on the page focus on healthcare, where a clinician may need to review why a patient-facing assistant surfaced a medication, allergy, or past visit, and on enterprise agents that make decisions based on remembered business or customer information. It is also presented for startups and small teams building agent products that need memory across sessions and multi-tenant isolation.

The service is live via API today, and Python and JavaScript SDKs are available. The page also references pip install aurra, and says native, database-enforced tenant isolation is part of the system. Pricing includes a free start, with “Start free” shown on the page and a “Pricing” link available. Aurra is a memory layer for AI agents rather than a general-purpose application, and it emphasizes verifiable, bi-temporal, multi-tenant agent memory.