Tenant access boundary
Users see only their licensed workspace, assigned projects, and role-allowed actions. Superadmin screens manage organizations, licenses, and users without displaying tenant audit posture or evidence content.
Privacy and security model
Evidence stays controlled. AI sees only the minimum needed context.
EvidPrism is built for regulated audit work where evidence confidentiality, tenant access control, human review, and traceable output matter as much as speed.
Evidence processing choices
Recommended mode lets the server parse evidence for stronger control mapping and citations. Client-encrypted raw-file mode encrypts raw files in the browser and stores only ciphertext plus best-effort redacted search text.
Evidence Privacy Gateway
Cloud AI routing uses retrieved evidence snippets after best-effort pattern-based masking, reversible tokenization, residual-risk checks, and route ledgering. Raw files are not sent to the AI provider in the standard SaaS route.
Integrity trail
Uploads and key audit actions are hash chained. Customers can export or externally anchor the head hash for independent proof that the evidence ledger was not silently altered.
Human review gate
AI output is treated as draft audit assistance. Auditors review sufficiency, citations, observations, and exceptions before issuing professional output.
Deployment flexibility
The same codebase can run as SaaS, enterprise private deployment, or laptop/desktop deployment. Private model routes can be enabled where a customer requires local LLM processing.
Important limitation
Redaction is a control, not a magic promise.
Best-effort redaction reduces exposure before AI processing, but regulated customers should still approve evidence classes, AI routes, retention settings, and deployment mode before production use.