Security for legal work that needs a clear chain of trust.
Grella protects customer content with encryption, per-organization isolation, role-based access controls, and transparent AI-provider processing.
Current posture
Encrypted app data
Organization-scoped workspaces
AI-provider processing disclosed
No zero-knowledge, SOC 2, full file-access logs, or default enclave processing claim today.
Architecture ledger
The security story is concrete, not absolute.
Grella protects legal workspaces through layered controls: encryption, KMS-managed keys, organization isolation, and user permissions. It does not pretend those controls remove every operational access path.
Encrypt
Sensitive fields use AES-256-GCM.
Chat messages, artifact content, extracted facts, and document chunks are encrypted as sensitive application data.
Manage Keys
Per-organization keys are managed with Google Cloud KMS.
Grella uses an envelope-encryption pattern with organization-specific keys and version tracking.
Isolate
Workspace data is scoped by organization.
Repository queries are scoped by organization, and file deduplication is per organization instead of global.
Control
Roles and matter permissions decide access.
Seven roles, granular permissions, per-matter access, and chat visibility controls define who can see what.
Data movement
From upload to answer, the content path stays visible.
This is the honest flow behind Grella: files are stored and indexed, relevant passages are retrieved, and AI providers process the context needed to answer.
Upload
Files move over HTTPS/TLS into cloud storage.
Extract
Text is OCRed, chunked, embedded, classified, and prepared for search.
Retrieve
Grella finds the passages relevant to the user question or generated work.
Process
Needed context is sent to business API AI and document-processing providers.
Review
The user reviews citations, facts, and drafts before relying on the output.
AI provider truth
Grella says exactly what leaves the workspace.
Document text and queries can be sent to AI and document-processing providers. The security promise is no training on customer API data, not zero provider access.
Read the privacy detailsWe say
AI providers process document text and queries to provide Grella AI features.
Business API providers state they do not train models on customer API data.
Customer content is not used to train Grella-owned foundation models or build models for other customers.
We do not say
We do not say AI providers never technically receive customer content.
We do not claim signed DPAs with every AI provider today.
We do not redact all PII before provider processing today.
Logging is useful, but not universal yet.
The page distinguishes current work-product logs from broader file and search audit logs that are still planned.
Implemented
Artifact and work-product audit logs
Create, edit, access, reference, export, restore, and delete events are logged for generated work product.
Limited
Recent resource access
Some recent access is tracked for product navigation and operations, not as a complete compliance audit log.
Not claimed
Every file view and search
Grella does not currently claim full logging for every raw file view or every search query.
Planned
Broader file/search logs
Broader file and search access logging is planned as the platform matures.
Current, optional, not claimed
Security posture without procurement theatre.
These groups keep baseline controls separate from advanced enterprise discussions and roadmap items.
Current
AES-256-GCM on sensitive fields
Google Cloud KMS
Organization isolation
Role-based access controls
Artifact audit logs
Optional / Scoped
TEE or enclave-style processing
Cloud HSM-backed keys
Dedicated infrastructure discussions
Data-residency discussions
Customer-specific terms
Not Claimed Yet
SOC 2 or ISO certification
Full file/search audit logs
DPA template
PII redaction before all AI calls
Security monitoring program