Support Built for Busy Small Firms

Very easy. The installer runs with one double-click and takes about 2 minutes. A setup wizard walks you through everything: choosing your AI backend, connecting to AWS Bedrock (or setting up local AI), and processing your first test case. Most attorneys set it up themselves without any help. If you have an IT provider, they'll find the deployment straightforward — it's a standard Windows installer with a local web interface. Either way, you're up and running the same day.

MedRecords AI runs as a local web application. Once installed on any computer in your office, your paralegal (or anyone on your network) can access it through their web browser by navigating to the app's URL. There is no separate login for each user — the software is protected by a single password you set during setup. For multi-computer setups, contact us about multi-seat licensing.

Yes. At 15 cases/year, you're currently spending 90–120 hours on manual record review (6–8 hours per case). At a $300/hr billing rate, that's $27,000–$36,000 in lost revenue opportunity. MedRecords AI costs $4,500 once and cuts that time by 90%. You'll see ROI after your first 2–3 cases. Plus, the time you get back lets you take on additional cases you'd otherwise turn away.

A full-time paralegal costs $45,000–$65,000/year, takes 3–6 months to fully train, and maxes out at about 20 active cases. MedRecords AI costs $4,500 once (no recurring fees for the software), requires zero training beyond a 2-minute setup, and helps your existing paralegal handle 3–4x more cases. It doesn't replace human judgment — it eliminates the hours of manual reading so your team can focus on strategy and client service.

Download the installer from your purchase confirmation email and run MedRecordsAI-Setup.exe. The installer handles everything automatically — Python, dependencies, and application files. Once installed, double-click the desktop shortcut or launch.bat and follow the built-in setup wizard. The wizard configures your AI backend (AWS Bedrock or local Ollama), initializes the database, and walks you through your first case.

Windows 10 or later, 8 GB RAM minimum (16 GB recommended if running local AI models via Ollama). Python and all dependencies are installed automatically during setup. If you choose local AI (Ollama), it will also be installed and configured automatically. The application runs entirely in your browser at localhost.

The same one. Pro and Demo share a single installer — the license key you receive by email after purchase unlocks the Pro feature set on any existing install. Download the installer from our homepage, run it, and then paste your key in Settings → License. No reinstall, uninstall, or separate Pro download is required. If you had the Demo running already, activation is instant and non-destructive — all your data and settings are preserved.

After launching MedRecords AI, navigate to Settings → License and paste your license key into the activation field. Your key was sent to the email address used during purchase. The license activates instantly — no internet connection is required for validation.

Each license covers one machine. If you need to run MedRecords AI on multiple computers, contact us at dan.direnfeld@aiproductivity.dev for multi-seat pricing. We offer volume discounts for firms with multiple workstations.

No. As of v2.2.6, MedRecords AI includes an Automated Dependency Manager. Once Ollama is installed and running, the application will automatically detect it and pull the required medical models (Qwen and GLM) in the background. You will see a "Downloading" status in the logs during the first run.

Nothing is lost. MedRecords AI uses a Resilient Background Worker with checkpoint-based auto-resume. If you close the browser, your computer restarts, or the process is interrupted, the pipeline automatically resumes from the last completed stage when the application restarts. Each stage has its own timeout with automatic retry, and a push-folder retry queue ensures completed reports are always delivered even if the output folder was temporarily unavailable. All progress is stored in the local database for maximum durability.

Yes. Files in the local vault (including summaries and audit logs) are encrypted using Fernet authenticated encryption (AES-128-CBC + HMAC-SHA256). Encryption keys are generated per installation and stored in the operating system's credential store (Windows Credential Manager / macOS Keychain), never co-located with the encrypted data — so an attacker who copies the vault files alone cannot decrypt them.

When you process a case, the text is indexed into a local Vector Store on your machine. When you ask a question in chat, the system performs a high-speed local search to find only the most relevant snippets. This data is processed locally by Ollama or sent securely to AWS Bedrock (if cloud AI is enabled) only when requested. No data is used to train external models.

MedRecords AI now includes an Automatic Self-Updater. You can check for updates directly from the Admin dashboard. If a new version is available, the app will download the patch and prompt you to restart to apply the latest security fixes and features.

Upload your document via the Files tab and click Summarize. By default, MedRecords AI uses Ledger mode to extract every medical event into a structured billing ledger with dates, providers, codes, and amounts. The pipeline runs through four stages (Ingestion, Privacy, Extract, Verify) with real-time progress tracking. Once complete, you can view the case briefing card, browse extracted items, generate demand packages, and run case reconciliation.

MedRecords AI supports PDF (both native text and scanned documents via OCR), TIFF, PNG, JPG, DOCX, DOC, XLSX, XLS, TXT, CSV, JSON, XML, LOG, HTML, and medical DICOM imaging files (.dcm, .dicom). Scanned PDFs and image files are automatically processed through the built-in OCR engine with medical terminology correction. Maximum upload size is 100 MB per file.

Ledger (default, 5–12 min) is the forensic processing mode that extracts every medical event into a structured billing ledger with dates, providers, CPT/ICD-10 codes, and dollar amounts. Each item is individually verified by an accuracy critique before being written to the database. Fast (1–2 min) provides a quick narrative overview suitable for initial case screening. Auto (2–5 min) offers a balanced narrative analysis with timeline and gap detection. Thorough (5–15 min) performs deep narrative analysis with cross-referencing and comprehensive source linking. Ledger mode is recommended for demand packages, settlement calculations, and trial preparation.

Yes. MedRecords AI integrates with Ollama, allowing you to run open-source AI models locally on your own hardware. Popular options include Llama, Mistral, Qwen, and other models. This enables fully air-gapped processing where no data ever leaves your machine.

Your one-time license fee includes the complete MedRecords AI software, Standard Support (knowledge base, forum, FAQ, and email support from our team), and all features available at the time of purchase. There are no recurring charges for the software itself.

The only optional recurring costs are Priority Support ($2,000/yr) and AWS Bedrock usage if you choose cloud AI processing (typically $1–$2 per case). All cloud processing goes through HIPAA-eligible AWS Bedrock, covered by the AWS Business Associate Agreement. You can avoid all recurring costs entirely by using Ollama for free local AI processing.

Priority Support ($2,000/yr) includes a 24-hour email response SLA, phone support during business hours, a 4-hour critical issue response guarantee, all version upgrades, a dedicated onboarding call, and quarterly check-in calls. It is especially valuable for firms where downtime directly impacts case deadlines. At just $167/month, it's like having an on-call IT team.

MedRecords AI installs on your firm's hardware. AI inference runs in one of two customer-controlled paths: locally via Ollama (air-gapped — no data leaves the workstation) or through your firm's own AWS Bedrock account under your firm's own AWS Business Associate Agreement (PHI is transmitted to AWS over TLS using the AWS SDK's defaults; AWS Bedrock does not retain inference data or use it to train models). Productivity AI is not in the PHI data path. Local data is encrypted using Fernet authenticated encryption (AES-128-CBC + HMAC-SHA256), and an HMAC-signed audit log records every access and action with six-year retention.

For AI inference, only if you use AWS Bedrock. With Ollama configured as the inference backend, no inference data leaves the workstation. A few small calls remain unless explicitly disabled in Settings: an initial one-time license activation, a periodic license check-in (every ~30 days, with a 60-day offline grace period), an optional update check, and optional anonymized usage telemetry (no PHI). All of these can be blocked at the firewall for fully air-gapped operation.

MedRecords AI is designed to support a HIPAA program through on-premise application architecture with a customer-controlled AI inference path (Local-Only Ollama or BYO-AWS Bedrock under your firm's own AWS BAA), Fernet authenticated encryption (AES-128-CBC + HMAC-SHA256) for vault files, immutable audit logging (SQLite triggers prevent deletion or modification of access-log entries), soft-delete data retention (6-year retention with automated purge), automated breach detection (alerts on bulk view/download/export patterns), PHI-sanitized error messages, and role-based access controls. When using AWS Bedrock, a BAA acknowledgment step is built into the setup wizard. Productivity AI does not operate a cloud AI endpoint that receives your PHI under either deployment, so the traditional cloud-vendor BAA chain is reduced to your direct BAA with AWS (or eliminated entirely in Local-Only mode). HIPAA compliance remains the licensee's responsibility; the Software is one component of your program.

The installer automatically detects and installs Python if it's missing. If you still encounter issues, run the application from a command prompt to see error messages. Common issues include port 8801 already being in use by another application — check with netstat -aon | findstr :8801 and close any conflicting processes. MedRecords AI automatically kills zombie processes on startup, but you can also manually clear the port. If Python was just auto-installed, you may need to close and reopen the command window for PATH changes to take effect.

Download the latest version from your account or purchase confirmation, then extract the files over your existing installation directory. Your data, settings, and license key are preserved automatically — only the application code is updated. No reinstallation or reconfiguration is needed.

Ensure there are no extra spaces before or after the key when pasting. If the issue persists, contact support at dan.direnfeld@aiproductivity.dev with your order confirmation and we will resolve it promptly.

MedRecords AI handles large cases with per-stage timeouts and automatic retry. If any stage times out, click Retry — the pipeline resumes from the last completed stage, so no work is lost. On restart, the application also auto-resumes any interrupted pipelines (up to 3 attempts). For very large documents with hundreds of extraction items, the Verify stage may take several minutes as it batches accuracy checks — this is normal and shown in the progress bar. If timeouts persist, try Thorough mode for narrative processing.

A pipeline watchdog runs in the background and automatically detects stuck pipelines. Each processing stage has its own timeout — if any stage exceeds its limit, it is safely terminated and the pipeline is marked as failed with a Retry button. On restart, the application automatically resumes interrupted pipelines from the last completed stage (up to 3 attempts). If the issue persists after retry, run medcli system doctor --fix to diagnose and auto-repair common issues.

MedRecords AI is designed to handle cases of any size, including 2,000+ page records across dozens of files. Large cases are automatically processed with adaptive timeouts that scale with document size, partial-progress recovery on transient failures, and a Retry button that resumes from where it left off rather than starting over. The progress bar tracks each stage end-to-end so you always know where the run stands.

Negligence Detection (Pro) analyzes a processed case for standard-of-care deviations. It identifies timing rule violations (e.g., delayed ER imaging), medication contraindications, documentation inconsistencies, and treatment gaps. Each finding includes a severity level, citation to source records, and a recommendation. You can accept or reject individual findings with one click, and your feedback is stored for continuous improvement.

Auto Task Generation (Pro) reads your completed summary and creates a prioritized list of action items — depositions to schedule, records to request, experts to retain, follow-ups needed, and more. Tasks are sorted by five priority levels (Critical through Routine) with suggested assignees (Attorney or Paralegal). You can export the full task list to CSV for import into your practice management system.

Yes. The Deposition Summarizer analyzes uploaded deposition transcripts to extract key testimony, identify contradictions, flag admissions, and cross-reference statements against the medical records. Upload a transcript through the Depositions tab to get a structured summary of the most relevant testimony.

Yes. Download the pre-formatted Excel template from the Contacts page, fill in your contact details (medical providers, insurance companies, etc.), and use the batch import feature to upload them all at once. Imported contacts integrate with the Email Retrieval Bot for automated records request follow-ups.

The Records tab (v2.5.0) is a dedicated records collection management center with three sub-tabs. Contacts lets you manage all medical providers, custodians, and facilities with sortable tables, filters, batch import/export, and one-click follow-ups. Cases groups contacts by case reference with visual progress bars showing collection status. Activity provides a chronological timeline of all email correspondence for full audit visibility. All contact data is encrypted at rest and every access is logged for HIPAA compliance.

When you add a contact in the Records tab, MedRecords AI automatically sends an initial records request email. If no response is received, you can send tiered follow-ups (polite, firm, final) with one click from the contact detail modal. The system tracks follow-up counts, last contact dates, and collection status for each provider. You can also follow up all pending contacts for an entire case at once from the Cases sub-tab.

The AWS Wizard (v2.5.0, found in Settings) walks you through connecting MedRecords AI to AWS Bedrock for cloud-based AI processing. Enter your AWS credentials (ABSK Bearer Token or Access Key pair), and the wizard validates your connection, verifies model access, and securely stores your credentials in the system keyring. No manual configuration files needed.

The Virtual Paralegal (v2.6.0, Virtual Paralegal tab) is an AI agent that executes real commands on your behalf through natural language chat. Tell it what you need — "check pipeline status," "follow up on pending records for case 2024-Smith," "generate a demand package" — and it translates your request into the correct CLI operations. Safe read-only commands execute instantly; write operations (sending emails, modifying settings) always require your approval first.

Watch folders are directories that MedRecords AI monitors continuously. Any new PDF or document dropped into a watch folder is automatically ingested into the processing pipeline — no manual upload required. Push folders receive completed summary reports automatically. You can configure both through the Virtual Paralegal (paste a path and it asks "watch or push?") or via CLI commands like medcli folders add-watch --path "C:\Cases\Incoming".

MedRecords AI includes a built-in System Doctor that runs 10 automated health checks (database integrity, port availability, API connectivity, disk space, and more). Run medcli system doctor from the command line to diagnose issues, or medcli system doctor --fix to auto-repair common problems. For severe issues, medcli system nuclear-reset --confirm performs an emergency reset (kills zombie processes, clears locks, resets state). You can also ask the Virtual Paralegal to "run system doctor" from the Virtual Paralegal chat.

Ledger Mode (v2.10.0, now the default) is a forensic-grade processing pipeline that extracts every medical event into a structured billing ledger. Instead of producing a narrative summary, it creates individual line items with dates, providers, CPT/ICD-10 codes, billed amounts, and source citations. Each extracted item is then verified by an AI accuracy critique. The result is a comprehensive, auditable medical billing ledger that powers demand packages, settlement calculations, and case briefing cards. Ledger mode processes files through four stages: Ingestion, Privacy, Extract, and Verify.

Accuracy Verification (v2.11.0) is an automated quality gate that runs after extraction in Ledger mode. Every extracted item is checked for source grounding (does the original document support this claim?), factual accuracy, and citation validity. Items that fail verification are flagged with reduced confidence scores or marked for review. For documents with many items, verification runs in batches and typically takes 1–6 minutes. The progress bar on the Summarization page shows this as the "Verify" stage.

Case Reconciliation (v2.11.0) is a per-case quality audit that runs automatically after all documents in a case are processed. It performs five deterministic checks — billing consistency, timeline coherence, diagnosis alignment, provider cross-referencing, and citation validity — plus an LLM-powered narrative cross-check. Results appear as a green/amber/red badge on the case detail page with an expandable panel showing any discrepancies found. You can also trigger reconciliation manually from the case detail page.

The Case Briefing Card is a one-page summary that appears at the top of each case detail view. It aggregates data from all processed documents in the case to show total billed amounts, date range of treatment, primary injuries, treating providers, and key findings. The briefing card is generated automatically from the master case aggregator and updates as new documents are processed. It provides attorneys with an instant snapshot of case value without reading through individual summaries.

When you generate a demand package from a case, MedRecords AI pulls the master billing ledger to create a Master Ledger Summary table directly in the DOCX output. This includes every provider, every visit, billed amounts, and ICD-10 codes — all sourced from the verified extraction data. The demand package also incorporates the case briefing card data for the demand summary section, ensuring the dollar figures in your demand letter match the verified medical billing records.

The audit log is protected by immutable database triggers (v2.10.0) that prevent any DELETE or UPDATE operations on audit entries. Even direct database access cannot modify or remove logged events. Additionally, audit entries include HMAC signatures that detect any external tampering. This ensures a complete, unalterable chain of custody for every access, export, and modification to medical records — critical for HIPAA compliance and litigation defensibility.

v2.31.0 is a stability + reliability release that hardens the production path against a batch of edge-case bugs found during partner testing. Highlights: worker-first license activation so post-demo upgrades work portably across machines (no more "Invalid license key format" on a valid key); cancel/retry policy fix — user-cancelled pipelines no longer auto-resurrect on restart, force-cancel and watchdog-timeout no longer release the slot while the worker thread is still alive (preventing duplicate concurrent LLM work); commit-before-response — database commits now run in after_request so a transient SQLite lock can no longer masquerade as a 200 OK; streaming uploads + backups — secure-upload and full-backup paths stream to disk in chunks instead of loading whole files into memory; IMAP retrieval schema fix — expected_receipt_date + last_reply_text columns added so the inbound-email loop stops hitting "no such column" and reprocessing the same message; port-killer guard — startup will only terminate MedRecords / Python processes on the configured port, never an unrelated service. Earlier (v2.30): 2,000+ page record sets, DOI-anchored accuracy, mail-merge template engine in the Sales campaign wizard, unified Demand Package filenames, Case Pros & Cons rename, rebalanced ForensicGuard score.

Upgrade to v2.31.0 or later. Earlier Sentinel builds rebuilt their ambient context on every message, which added 1–10 seconds of overhead per turn. Current builds use a 60-second shared cache that reuses the context across chat messages in the same session, and cap any single LLM call at 60 seconds so a slow provider response no longer blocks the browser. If you are already on v2.31.0 and still see slow responses, check the Settings → LLM page to confirm your AWS Bedrock or Ollama backend is healthy, and look at logs/medrecords.log for Bedrock circuit-breaker messages.

The health page reports degraded when at least one non-critical subsystem is unhealthy — commonly no AI credentials configured, IMAP/SMTP not reachable, or endpoint encryption status unknown. The app continues running; features depending on the affected subsystem will return a clear error rather than silently fail. Click System Doctor on the Admin page to see exactly which checks failed and follow the suggested fix for each. Run medcli system doctor --fix from the command line for auto-remediation of common issues.

You’re running with the default admin username. Go to Settings → Account and set a unique username that isn’t one of the common defaults (admin, administrator, root, user). MedRecords AI surfaces this warning to flag a common credential-stuffing vector. Changing the username takes effect immediately and invalidates any active sessions.

The app tries to detect BitLocker status via Windows’ manage-bde tool. On some Windows editions (Home, or some Pro variants without the BitLocker feature enabled), the output can’t be parsed and the status returns unknown. That does not mean your data is exposed; it means we can’t confirm the disk is encrypted automatically. Verify in Control Panel → BitLocker Drive Encryption that the drive holding MedRecords AI is turned on. For MSP-managed deployments, we recommend BitLocker for the drive as a standard control regardless of detection status. The Vault layer also encrypts PHI at the application level using Fernet authenticated encryption (AES-128-CBC + HMAC-SHA256), so your records are encrypted at rest by MedRecords AI even if BitLocker is off.

Yes, up to 5 consecutive crashes with exponential cooldown. The launch_pro.bat wrapper monitors exit codes and restarts MedRecords AI automatically on non-zero exits. Each crash is logged with a trace ID to logs/crash_trace.log. If crash count reaches 5 within a window, the wrapper stops and shows a diagnostic message — at that point, open the log file, look for the most recent [crash] entries with their trace IDs, and share with support. Delete the .crash_state.txt file to reset the counter after fixing the underlying issue.

No. MedRecords AI is a desktop software product that runs on your firm’s own hardware. We (the vendor) never receive, store, or transmit your clients’ PHI — your records stay on your workstation. You do not need a Business Associate Agreement with MedRecords AI the company, the same way you don’t sign a BAA with Microsoft to use Word. The BAA you need is with AWS, because when you enable cloud AI processing, your records are transmitted from your workstation to AWS Bedrock using your AWS credentials. AWS provides a free, self-service BAA for HIPAA-eligible services through the AWS Artifact console. Execute that once per firm and you’re covered. If you use Ollama for fully local AI instead of AWS Bedrock, no BAA is needed at all because no data leaves your hardware.

Log in to your AWS account, open AWS Artifact → Agreements → Business Associate Addendum, and accept online. It’s free, takes about 5 minutes, and only needs to be done once per AWS account. Once accepted, all HIPAA-eligible services in that account — including Bedrock — are covered. During MedRecords AI’s setup wizard, the BAA acknowledgment checkbox asks you to confirm you’ve completed this step; the confirmation is logged to your compliance audit trail.

Nothing from your cases. Our servers see: license-key activation events, anonymized crash trace IDs (if you opt in to crash reporting), and aggregate usage telemetry (e.g., “case processed in ledger mode,” never the contents). Your PHI never touches our infrastructure. The setup wizard, processing pipeline, Virtual Paralegal, demand-letter generator, and every other feature runs locally on your workstation. Outbound connections go to: AWS Bedrock (your account, your PHI path), DigiCert for signed-update verification, our update manifest at aiproductivity.dev (no PHI transmitted, just version checks), and optional SMTP/IMAP servers you configure for records retrieval.

The breach monitor watches the audit log for anomalous patterns — unusually high volume of PHI views, bulk downloads, mass exports, or mass deletions by a single user within a short window. When a threshold is crossed, an alert is written to the breach_alerts table, flagged on the Compliance page in-app, and shown to any user with admin scope on next login. The alert includes the user, pattern, count, and time window. You review, acknowledge, and document the resolution; all of this is captured in the audit log. A circuit breaker prevents breach alerts from cascading during legitimate high-volume operations (e.g., an attorney bulk-exporting records for a production request).

When you delete a contact, case, or document, it’s soft-deleted — marked with a deleted_at timestamp but kept in the database for the HIPAA-mandated 6-year retention window. After 6 years from the last access, the automated purge job removes soft-deleted records permanently. If you need to force an immediate hard delete (e.g., at a client’s request under state privacy laws), contact support and we’ll walk you through the documented exception path. All retention and purge events are logged to the audit trail.

Yes. Go to Compliance → Audit Log and export to CSV or JSON. Filter by date range, user, action, or resource type. Each entry includes timestamp, user, action, resource, IP address, and (where applicable) the document or case affected. Exports are themselves logged for audit completeness. For a turnkey compliance report covering the last quarter — who accessed what, which breach alerts fired, audit integrity checks — use Compliance → Generate Report.

The text needed for the specific task — for extraction, summarization, chronology construction, demand letters, valuation, deposition outlining, and chat — is transmitted to AWS Bedrock in its original (un-redacted) form. Authentic PHI is required for the product to produce accurate clinical findings and citations. The cloud path is covered by your firm's own AWS Business Associate Agreement: Bedrock transmits via TLS 1.3, processes the request, and is contractually prohibited from retaining your inputs or training on them under the AWS BAA. An optional cross-model verification step may transmit the generated summary to a secondary AI model family under the same AWS BAA for accuracy review. If you want zero data to leave your hardware, configure Ollama as your backend instead: the same pipeline runs against local models (Llama, Qwen, Mistral, GLM) and no outbound network traffic occurs.

Once per month, MedRecords AI performs a quick license heartbeat to confirm your Pro license is active. This piggybacks on the telemetry reporter and transmits no PHI. If you’re offline, the app continues working on a 60-day grace period. At day 30 offline you’ll see a reminder; at day 60 the app asks you to reconnect to validate before processing new cases. Existing processed cases remain fully accessible in any state.

The app checks https://aiproductivity.dev/dist/manifest.json once every 24 hours. If a newer version is available, the update is downloaded over HTTPS and Authenticode-verified against Productivity.AI, LLC’s DigiCert EV code-signing certificate before being applied — so a man-in-the-middle cannot inject a rogue update. You’ll see a prompt before the update applies; your data, configuration, and license key are preserved across updates. You can disable auto-update in Settings → Update if your MSP mandates manual update windows.

Yes. Email dan.direnfeld@aiproductivity.dev with your order number and we’ll reissue a key bound to the new machine. The old installation is automatically deactivated once the new key activates. We don’t charge for legitimate hardware refresh or device replacement.

Settlement Estimation uses the verified case ledger (billing totals, injury severity, treatment duration, provider type, diagnoses) plus jurisdiction data to produce a suggested settlement range. The estimate is not legal advice and not a substitute for attorney judgment — it’s a data-grounded anchor. Each estimate includes a confidence band, the factors that moved it up or down, and citations back to the specific ledger items. Click Recalculate after adding new records; the estimate updates incrementally with the ledger.

Nothing for end-users directly — it’s an admin dashboard for firms managing their own customers (e.g., an MSP running MedRecords AI for multiple law-firm tenants). It shows customer health scores, onboarding status, usage trends, and renewal windows. Non-MSP customers can ignore the Customer Success tab entirely; it doesn’t affect your day-to-day workflow.

MedRecords AI is designed for a single workstation with one active user session at a time. If you need two paralegals both processing cases simultaneously, the recommended setup is one MedRecords AI install per workstation. Current builds (v2.31.0+) handle concurrent write operations — contact imports, bulk file updates, etc. — reliably, with 20 simultaneous record-creation operations completing in under 3 seconds. But concurrent logins aren’t a supported deployment model; for multi-user firm workflows, talk to us about site licensing.