Trust Center: Technical Details

This is the detailed version of the Trust Center, written for IT reviewers and for AI assistants evaluating DocketDrafter on a firm's behalf. It covers DocketDrafter for Claude and ChatGPT. Firms that draft with DocketDrafter playbooks under a signed services agreement are covered by that agreement. Every claim on the main page is stated here with its specifics. If you need a detail that isn't covered, email support@docketdrafter.com.

Related pages:

• Trust Center, the plain-language version for attorneys
• Subprocessors. A firm that asks (support@docketdrafter.com) is notified before a new subprocessor starts handling its customer content.
• Terms of Service: §6, content and purposes; §7, confidentiality; §8, no training; §9, access and providers; §13, retention/deletion/export; §17, security and incident notice
• Privacy Policy

Company

DocketDrafter is operated by Avize LLC d/b/a DocketDrafter. The founders, Tommy Eberle and Grant Willard, build and run the service themselves. There are no employees or individual support contractors with access to customer content; the service-provider companies below also process data.

ToS §7 treats customer content (including drafts, uploads, instructions, outputs, client and matter details, and client information in feedback) as confidential. Sections 6–9 define permitted service purposes: providing, configuring, implementing, maintaining, supporting, troubleshooting, testing, assuring quality, operating, and securing the firm's service, fulfilling an order, enforcing the agreement, and complying with law. Identifiable content used for testing must be reasonably connected to delivering or supporting that customer's service, not unrelated research, general-purpose test datasets, demonstrations, or marketing. Generalized know-how and non-identifying analytics are addressed separately in the agreement.

These restrictions continue while content is retained, including after cancellation. Necessary disclosure to providers under confidentiality and data-use obligations, confidential professional advice, and legally compelled disclosure are addressed in the Terms. The Subprocessors page describes provider processing. The Terms define contractual obligations and limits on remedies; implementation details here are informational unless expressly incorporated into an agreement.

Architecture

DocketDrafter is a document workspace that the firm's AI assistant (Claude or ChatGPT) connects to through an MCP connector. The assistant does the reasoning and writing. DocketDrafter stores the firm's files, runs the assistant's commands in an isolated sandbox, and renders files such as Word documents, PDFs, and court forms.

DocketDrafter does not call AI models while the firm's assistant uses the service. DocketDrafter's own AI processing happens through the founders' tools for setup, maintenance, and support, described below. Firms that draft with DocketDrafter playbooks are governed by their signed services agreements.

The firm's assistant is governed by its agreement and settings with that provider. DocketDrafter does not receive conversation history as a separate chat feed. It receives tool submissions, including files, edits, commands, results, and feedback, which can themselves contain conversation excerpts. Separately, founders' AI tools can process workspace content during service work as described below.

Data storage

All traffic to and from DocketDrafter uses TLS. The storage bucket refuses unencrypted connections, and its public access is fully blocked. The workspace storage locations above are in the United States. Primary storage location does not establish the location of every access, founders' local working copy, or downstream processing operation. The provider list distinguishes storage from other processing. Encryption of sandbox working copies and other provider systems is governed by those providers' controls; see the provider list. For customer workspace work through Baseten, we use only GLM 5.3 Flash. Baseten support confirmed that this model is currently hosted in the United States. This is a model-specific hosting confirmation, not a claim of account-wide US-only routing across other models.

Sensitive identifiers that official forms require (Social Security numbers, A-numbers, dates of birth) are stored with the same protections as other workspace content; there is no separate identifier vault.

Suitability and authorized information

Confidential legal information is an intended use, including SSNs, A-numbers, identity documents, immigration status, sensitive personal narratives, and information about minor clients. Sensitivity alone does not require separate approval. The service is for adult legal professionals, not direct use by children.

The firm must determine that our disclosed processing is permitted by applicable law, court orders, client instructions, and contracts. Restrictions need not be absent, but must be compatible with the service. Requirements beyond standard commitments need express written agreement before submission. Classified information and information whose requirements the service cannot meet are not accepted. We don't offer HIPAA business associate agreements, so processing that requires one isn't supported; medical information alone does not establish that relationship. We do not inspect every document to identify undisclosed restrictions. These disclosures do not provide ethics clearance or guarantee privilege; the firm remains responsible for authorization, professional judgment, and output review.

Firm isolation

Each firm's workspace and stored files are private to that firm. A signed-in user and their assistant can reach only the workspaces they belong to. The founders and disclosed providers also process workspace data.

Every member with access to a workspace can access every matter in it. There are no private matters, per-matter permissions, or ethical walls within a workspace. Firms needing segregated access must arrange separate workspaces with appropriately separated memberships or keep affected work outside the service. Separate workspaces do not limit founder or maintenance-AI access.

The assistant's sandbox

The assistant's commands run in an isolated sandbox (an E2B microVM) holding a working copy of the firm's repository. Each user has their own sandbox, kept between sessions so work resumes quickly.

• Restricted network access. The sandbox's only network access is to DocketDrafter's package registry (packages.docketdrafter.com), which serves form and template packages. The registry accepts only authenticated GET and HEAD requests. Commands cannot freely email, upload, or post files to outside sites.
• Document downloads return to the firm's assistant, through signed links expiring after one hour. Results and feedback also pass through DocketDrafter's service; disclosed providers process data.
• No user-facing external-sharing, email-sending, publishing, or court-filing features. This is not a guarantee against onward disclosure through the firm's assistant or other connectors. Those permissions are controlled separately by the firm.
• Sandboxes are deleted after 30 days without use. A sandbox no longer linked to a user is deleted within two days.

Version history

The workspace is a Git repository. Version history supports recovery of saved changes, but is not a guarantee against every loss or an audit of every document read. Permanent deletion requests remove the requested history too.

• Every change is committed automatically when the assistant's command finishes, even if the assistant doesn't commit explicitly. Each version records when it was made, which user's session made it, and a description.
• Deleting a file doesn't erase it. It stays in the history and can be restored.
• History rewrites are backed up. If an assistant rewrites history, the previous state is kept as a recovery point automatically.
• Any two versions can be compared, and any earlier version can be restored. A Word redline can be produced for supported document formats and comparisons.
• Storage is versioned. An overwritten or deleted stored file's previous copy is kept for 7 days after becoming noncurrent. This is distinct from Git history.

Access by DocketDrafter

Only the two founders have personnel access to customer content. There are no employees or individual support contractors with access; disclosed provider companies also process content. Before anyone else receives personnel access, every customer, including free-trial users, gets at least 30 days' notice identifying the role, the access involved, and the applicable confidentiality obligations. During that period a firm can cancel and request export and deletion; content subject to a pending deletion request is not accessed by new personnel, content subject to a pending export request is not accessed by them until the export is delivered, and content of a firm that cancels during the notice period is not accessed by them unless the firm later resumes a subscription (ToS §18).

The founders can access entire workspaces, including client documents and history, for the service purposes defined above. There are no matter-level or task-scoped technical restrictions on that access. We proactively review workspace activity, including commit messages, commands and their output, errors, feedback, and the documents involved, to identify problems and improve the firm's configuration, particularly during onboarding. We do not review client documents for purposes outside the service purposes. Access is from work laptops with FileVault, automatic locking, and MFA on accounts reaching customer data.

Local working copies. While actively supporting a firm, the founders may keep working copies of its workspace on those laptops, including the local session logs our AI tools keep. Copies are removed within 30 days after active support of the firm's workspace ends or the subscription is cancelled, and in any event as part of an authorized deletion request.

AI tools. We use coding agents and other AI tools backed by inference providers, currently including Anthropic, OpenAI, and Baseten, to configure, maintain, support, and operate workspaces. The tools can explore the entire workspace, including documents and version history. Access is not restricted to files selected for a task. Which agent interface or harness is used does not change our confidentiality, permitted-purpose, or no-training commitments.

Audit boundaries. Workspace history records saved changes. AWS repository-access events record downloads and pushes under the founder's identity and are retained for 365 days. These are not comprehensive logs of document reads or every input sent to a maintenance AI. We do not provide document-read auditing.

See Subprocessors.

AI training

DocketDrafter doesn't train AI models on customer content and uses its AI providers only under terms and configurations that do not permit training on it. Anthropic and OpenAI are used only through business-plan accounts; Baseten's Model APIs are used under contractual no-training and zero-data-retention terms. This applies regardless of the coding-agent harness or inference interface used. Model inference for service work is processing, not training. The firm's own assistant is governed by its separate agreement and settings with its provider; our commitments about providers we select do not configure the firm's account.

Feedback reports

The firm's assistant can send reports automatically or at the firm's request about bugs, missing features, confusing behavior, or product requirements. The tool asks for actionable detail, potentially including relevant document excerpts or matter context, while discouraging entire documents or conversations. Reports are linked to the user's account, stored in S3 with firm records, and posted to a private founders-only Slack channel. Slack copies expire after 30 days; S3 reports remain until deletion is requested. Client information in reports remains confidential Customer Content, distinct from reusable product suggestions.

The feedback tool is optional. A firm can disable it in its assistant's connector settings, or set it to require approval before each report is sent. Disabling it does not disable command/result records, other normal service activity, or founder access.

Calls with us

Before each call, we tell participants that we transcribe calls for our internal notes, in the booking confirmation and the reminder email. Anyone can tell us they'd prefer we don't, and we'll take notes by hand. We use Granola for internal customer records, organization, follow-up, and product feedback only. Model training is disabled in our Granola account. We do not retain audio or video recordings unless asked. Only the founders access retained transcripts apart from providers used to process and store them. Transcripts are not marketing content or public testimonials. Transcripts are kept as business records, with no fixed retention period. On written request to support@docketdrafter.com, we remove client and matter details from them.

Retention

Workspace source documents and their Git history are retained with the workspace. Generated-artifact expiration applies to rendered copies, not their retained source documents. Stored-file expiration in versioned storage can leave a noncurrent copy for the additional period above; an authorized permanent-deletion request removes applicable versions within the deletion deadline below.

Deletion and export

Cancellation does not automatically delete or export data. An authorized representative can request permanent deletion of the entire workspace or a single matter in writing at support@docketdrafter.com. This is a manual process; deleting a file through an assistant does not remove its history. Single-matter deletion includes associated history, uploads, reports, and working copies. We may ask the firm to help identify matter information in shared files or records involving multiple matters; the deletion commitment below still applies. Within 30 days of an authorized request, we delete the requested content, history, and associated copies (including applicable stored-file versions, sandboxes, reports, session records, and the firm's account records) from our systems, founders' devices (including local AI-tool session logs), conversation history in the founders' AI-provider accounts, and the providers that store it, and confirm completion in writing. Email correspondence and call transcripts are business records rather than workspace content; they remain confidential and are kept, and on written request we remove client and matter details from them. If law requires retention, we retain only the required information, continue to protect it, and explain the exception where legally permitted.

Some copies aren't edited and instead expire on their own:

• Operational and audit logs contain no customer content. They expire on the schedule in Retention.
• Database backups hold account records, never documents, and expire after 7 days.
• AI coding agents. What our coding agents sent their inference providers while we maintained the firm's workspace stays under those providers' terms: Anthropic deletes it within 30 days under its commercial terms for our Team plan, OpenAI retains it under its business terms for our ChatGPT Business workspace, and Baseten retains no Model API inputs or outputs. Conversation history stored in our provider accounts is deleted as part of an authorized deletion request; provider-side copies we cannot delete expire under the provider's terms.
• Billing records are kept as the law requires.
• Records of the firm's acceptance of our terms (who accepted which version, when, and from where) are kept as the record of the agreement.

The firm's conversations with its own assistant are held in its account with Anthropic or OpenAI, and the firm deletes them there.

During its subscription or after cancellation or termination, a firm can request an export at support@docketdrafter.com while the requested content remains retained. After verifying authority, we provide the latest version of each document rendered as Word (.docx) and PDF (court forms as fillable PDFs, with any continuation pages as Word files), the original files the firm uploaded that remain stored when the export is prepared, and the firm's context files (the instructions and conventions saved for the firm). Exports are not automatic. They contain current versions only, not version history; we do not promise to reconstruct deleted files or recover expired copies.

Exports exclude proprietary rendering tools, private registry access, infrastructure, credentials, and internal operational logs. Export does not transfer ownership of DocketDrafter Materials; secrets or material whose disclosure would violate law may be excluded.

Authentication and access control

• Sign-in is handled by Clerk. Users can turn on multi-factor authentication (an authenticator app, with backup codes), and a firm admin can require it for all members from the Team settings page, for both the web app and the connector's sign-in. Signing in from a new device without MFA requires a code sent by email.
• Connecting an assistant uses OAuth. The assistant gets a revocable access token, never the user's password. A firm can disconnect DocketDrafter from Claude or ChatGPT at any time.
• The founders' accounts for every system holding customer data (AWS, Railway, Supabase, Clerk, E2B, GitHub, Google Workspace, Slack) require multi-factor authentication.

Audit logging

A multi-region AWS CloudTrail trail records every AWS management API call in the account, including every use of the storage encryption keys, with log-file validation turned on. Logs are kept for 365 days in a dedicated, versioned, encrypted bucket. The trail records which systems used the keys and when, never document contents.

Incident response

We notify the firm's account administrator within 72 hours of confirming a security incident affecting its data, or sooner if applicable law requires. We describe what happened, affected data, and our response to the extent then known, and provide updates. This does not postpone earlier legally required notice.

To report a security concern, email support@docketdrafter.com.

Recommended setup for the firm's assistant and workspace

The firm's AI assistant is where its people handle client information, so its configuration matters as much as DocketDrafter's. These are recommendations, not eligibility conditions or a certification of ethical compliance.

In Claude or ChatGPT

• Use a business plan (Claude Team or Enterprise; ChatGPT Business or Enterprise), where conversations are not used for model training by default. On a consumer plan (Claude Free, Pro or Max; ChatGPT Free, Plus or Pro), disable the provider's model-training setting before submitting client information.
• Give each person an individual account, with MFA enabled.
• Disable connector operations the work doesn't need, such as sending email or deleting files, including in connectors other than DocketDrafter.
• Restrict network egress for the assistant's code execution and browsing to an allowlist of the domains the work needs, where the plan supports it (typically an organization-level admin setting).
• Disable or avoid conversation feedback (response ratings, bug reports) for chats containing client information. These features export the full conversation to the provider, which retains it under separate, typically longer, terms than ordinary chats.

In DocketDrafter and the firm

• Grant workspace access only to people who may see every matter in it. Use separately managed workspaces where access must be segregated.
• Require MFA for all members, from the Team settings page.
• Remove members promptly when they leave or no longer need access. Membership is re-checked on every connector request, so removal ends the member's assistant access immediately; no separate disconnect is needed.
• Verify source facts, translations, missing answers, form editions, and final outputs before relying on or filing them. Retain necessary client records independently.

The firm must determine client authorization and record-specific requirements and retain professional judgment.

What we don't have yet

• DocketDrafter doesn't have a SOC 2 report or a third-party penetration test. Our infrastructure providers (AWS, Railway, Supabase, Clerk, E2B, Axiom) each hold SOC 2 Type II reports, and Railway also publishes a third-party penetration test report. These reports cover the providers, not DocketDrafter itself.
• There are no private matters or per-matter access controls within a workspace.