Sidebar provides technical and contractual infrastructure designed to help attorneys use AI while maintaining their confidentiality obligations under ABA Model Rules and state bar guidance.
AI providers do not store your prompts or responses
All three model choices above run only under a signed zero-retention agreement — a model is never reachable in Sidebar until that agreement is in place. The certifications shown are held by the underlying infrastructure providers, not by Sidebar. Sidebar is an early-stage company and is not yet independently certified (SOC 2 is on our roadmap). Our own data handling is described plainly in the sections below.
How your data moves through the system
Your message travels from your browser to our server over HTTPS (TLS 1.2+). Your identity is verified via Clerk authentication.
Based on your selected model, we forward your message over a zero-retention channel: Claude (the default, running in an isolated zero-retention cloud environment), OpenAI with storage disabled on every request, or Google's Gemini under a signed zero-retention agreement. Every one of these is reachable in Sidebar only because that agreement is already in place — none is a bolt-on. The provider does not retain or train on your content.
The response streams back to your browser. The AI provider processes your request in memory and does not retain it.
Our usage log records which model was used, token count, estimated cost, and timestamp — not your message content.
Your conversations, uploaded documents, and research are saved to your organization's isolated database partition so you can return to them. Message content, extracted document text, research memos, and matter memory are encrypted with application-layer AES-256-GCM under Sidebar's own key, on top of infrastructure disk encryption. Uploaded original files and generated documents are kept in private, access-controlled storage. Workspace labels — conversation titles and matter names — are stored unencrypted so your workspace stays searchable (see Data Practices for the full breakdown). Deleting a conversation is a hard delete: messages, extracted text, and the uploaded original files are removed.
Each organization's data is separated at the database level
Every request re-checks organization and user ownership before any data is returned, so tenants are isolated at the application layer. Every database table also carries Row Level Security policies scoped by organization ID as a defense-in-depth backstop.
Powered by Clerk (SOC 2 Type II certified) with multi-factor authentication support, organization management, and role-based access control.
How Sidebar addresses the evolving landscape for AI in legal practice
The court held that a defendant's exchanges with a consumer generative-AI tool were not protected by attorney-client privilege or the work-product doctrine — in part because communicating through a public, third-party AI platform whose privacy policy permitted data collection did not support a reasonable expectation of confidentiality.
The reasoning turned on the tool's consumer data practices — retention and the absence of confidentiality commitments — which is precisely the posture that enterprise no-retention channels are designed to avoid.
Sidebar is designed to align with the enterprise characteristics the court distinguished:
The ABA's first formal ethics guidance on lawyers using generative AI requires lawyers to understand data practices, protect client information, and exercise supervision over AI tools.
Key differences in data handling and security
| Consumer AI | Sidebar | |
|---|---|---|
| AI-provider data retention | 30+ days | Zero |
| Used for training | May be | Never |
| Your saved conversations | Provider-controlled | You control — hard delete |
| Third-party disclosure | Per privacy policy | Contractually prohibited |
| Content encryption at rest | Varies | AES-256 (app-layer) |
| Encryption in transit | TLS | TLS 1.2+ |
| Tenant isolation | None | Per-request + RLS |
| Subprocessors | Partial | Vetted + published |
How uploaded documents are processed and stored
Uploaded originals and generated documents are stored in private, access-controlled storage scoped to your organization — never in any public bucket, and never shared across tenants.
Uploads are validated by magic bytes (file signatures), not just extensions. Only PDF, DOCX, TXT, and image files are accepted. Size limits are enforced (50MB documents, 10MB images).
Text extracted from your documents is stored in your organization's isolated database partition, encrypted at rest with AES-256 and protected by Row Level Security.
Document content sent to AI models follows the same zero-retention pipeline as chat messages. Providers do not store it and do not use it for training.
The vetted providers in our stack and the certifications they hold
Cloud model-hosting & compute
Claude inference (default) & document generation
Supabase
Database & storage (United States)
Clerk
Authentication & identity
Vercel
Application hosting
Certifications listed are held by each provider, not by Sidebar.
What Sidebar itself stores, and how you get rid of it
Your saved workspace: conversations, uploaded documents, research memos, and matter memory — the same way your practice-management or document tools store your files. AI providers keep nothing; Sidebar keeps what you choose to save, in your firm's isolated partition.
Conversation content, extracted document text, research, and matter memory are encrypted with application-layer AES-256-GCM under Sidebar's own key, on top of disk encryption. Uploaded originals and generated files live in private, access-controlled storage. Titles and matter labels stay unencrypted for navigation — use matter numbers instead of client names in titles if you prefer.
Deleting a conversation or matter is a hard delete — the messages, extracted text, and uploaded original files are removed, not hidden. Time-limited internal records (such as tool-audit content) are purged automatically after 30 days. Closing your account is self-serve full erasure: every record and stored file in your workspace is hard-deleted.
Civil demands for your content belong with you, the account holder — not with us. Unless legally prohibited, we notify you of any subpoena or demand before responding and give you the opportunity to object. Details in our Legal Process Policy.
This is also covered in our Privacy Policy and Legal Process Policy.
How we run the service, stated only where we can back it up
The application encryption key and service credentials are stored as managed secrets in our deployment environment, never in source code, and are held separately from the data they protect.
Internal diagnostic tooling that could reach customer data is gated behind an explicit control, never writes decrypted content to disk, and records each use in an access log.
Your workspace is backed up by our database provider within the same U.S. region and encryption boundary, for reliability and disaster recovery. Deletion propagates to these copies as they cycle out.
Changes go through code review and an automated pre-merge check suite before deployment. Schema changes are applied through versioned, reviewed migrations.
Report a vulnerability:if you believe you've found a security issue, email security@sidebar-ai.com. We investigate every report and will not pursue good-faith researchers who follow responsible-disclosure practices.
Sidebar provides technical and contractual infrastructure designed to support your firm's confidentiality obligations. We do not guarantee attorney-client privilege. Privilege determinations are made by courts based on the totality of circumstances. We recommend that firms using Sidebar obtain informed client consent for AI usage, independently verify all AI outputs, and establish internal AI usage policies consistent with applicable bar rules. This page is not legal advice.
We are happy to walk you through the details.