AI-Powered Document Compliance
Catch More.
Miss Less.
Cite Everything.
AI-powered compliance detection for regulated documents. Upload your norms. Submit your proposals, contracts, and filings. Catch what manual review misses — with a citation for every flag.
Book a DemoMeasured, not promised
Every labeled deviation caught, zero fabricated citations — ground-truth evaluation on real multi-document bundles, July 2026.
Why now
Regulators are shifting document review onto filers — SEBI's 2026 green channel makes fund launches self-certified. Pre-filing verification is the new mandate.
Agent-native platform
UI for reviewers, API for systems, MCP connectors for ChatGPT & Claude — live today.
Documents Deviate.
Buried clauses. Missing coverage. Data mismatches. Compliance gaps. Whether it's a security policy, an insurance proposal, or a Letter of Credit — deviations are caught too late, or not at all. Your team reviews hundreds. Every miss is a risk.
Agents You Can Build
Examples from regulated review work. Define yours.
InfoSec Policy Reviewer
Audits security policies against SOC 2, ISO 27001, and your control library — flags missing controls and clauses that don't comply.
Critical·MFA requirement absent from privileged-access policy
Health Claim Adjudicator
Reviews discharge summary, hospital bill, and diagnosis report together — catches cross-document inconsistencies and treatment-policy mismatches.
Critical·Discharge date precedes admission window
Fund Documents Reviewer
Checks offering memoranda (PPMs) against the regulator's prescribed template and fund rules — missing disclosures, fee inconsistencies, category mismatches.
Critical·Mandatory risk-factor disclosure missing
Trade-Finance LC Reviewer
Validates invoices, bills of lading, and packing lists against the Letter of Credit — currency, date, and description consistency.
Warning·Invoice currency differs from LC
Build for your domain.
Pick any document workflow where deviations cost real money. Define your norms. Train the agent. Ship it. The four above are starting points — your agent can look completely different.
How It Works
From raw norms to drafted action — the full loop your team walks, end to end.
Ground
Upload your regulatory documents, policy guidelines, and approved templates. Define the bundle contract — which documents make a complete submission. These become the norms your agent enforces.
Train
Review the deviations the agent flags on sample documents. Accept what's correct, reject false positives, challenge questionable rules. The agent's playbook gets sharper with every review.
Test
No agent goes live untested. Run it against a golden set of documents with known outcomes; an independent AI judge scores every run. The agent promotes to production only when it clears the gate.
Detect
Submit live documents — single files or multi-document bundles. The agent applies your norms and returns a deviation report, severity-rated, with the specific rule cited for every finding.
Communicate
The platform drafts follow-up messages from the deviation report — customer letters, broker notes, internal escalations. You review, edit, and send. Every draft is human-approved before it goes out.
Built for Precision
What makes DocTru different — the capabilities that separate an agent trained on your norms from a generic LLM prompt.
Agentic Deviation Detection
Not keyword matching. Not a generic LLM prompt. Each agent is trained on your norms — regulatory text, internal policy, approved exemplars — and learns from every review your team does. Flags carry severity ratings and cite the specific rule they violate, so every finding is auditable.
Measured, not promised: in ground-truth evaluation on real multi-document contract bundles, agents caught every labeled deviation — and fabricated zero citations. Internal evaluation, July 2026.
Multi-Document Bundles
Submit related documents as a single unit. The agent detects cross-document deviations — mismatched dates, missing artifacts, contradictions between docs — and cites every document involved.
Doc — Your AI Assistant
Ask anything about your agents, runs, training sessions, or platform usage. Doc reads your actual state, respects your role, and cites its sources.
Multi-LLM Consensus
Multiple AI models cross-check each other. Reduce false positives through agreement, not guesswork.
Learning Loop
Accept or reject flags. The agent learns. Precision improves with every review cycle.
Complete Audit Trail
Every scan, every flag, every norm citation. Inspection-ready compliance evidence.
Fits How Your Organization Works
UI for reviewers. API for systems. MCP for agents. Chat for everyone. One platform, reachable from wherever the work already happens.
UI for reviewers.
Web Application
The full review workspace — agents, training, submissions, deviation reports, analytics — role-gated for reviewers, agent managers, and admins.
API for systems.
REST API
A documented OpenAPI contract. Submit documents, poll runs, and pull deviation reports straight from your core systems and pipelines.
MCP for agents.
MCP Connectors
Add DocTru to ChatGPT or Claude as a connector. Submit a document in-chat and get the severity-rated report back — OAuth-secured, tenant-controlled.
Chat for everyone.
Doc, In-Platform Chat
Ask about your agents, runs, and usage in plain language. Doc reads your live state, respects your role, and cites its sources.
Your Data. Your Country.
Your Control.
Enterprise-grade security, built from day one.
Data Residency
Single-region by design. Data at rest never leaves your jurisdiction. AWS Mumbai today — additional regions deployable per customer.
No Training on Your Data
Your documents train your agents — never anyone's foundation models.
Encryption
AES-256 at rest. TLS 1.3 in transit.
Tenant Isolation
Row-level security. Database-enforced data separation.
Audit Trail
Every action logged. Every norm cited. Inspection-ready.
Let's Talk
Tell us about your compliance challenges. We'll show you how DocTru can help.
What to expect
- Response within 24 hours
- Personalized demo with your document types
- No commitment — pilot first, decide later