TrustAIby Bonis Systems
For AI-native legal firms

Anchor your AI's decisions
to a court-grade audit trail.

Your AI agents are making decisions a court, an opposing party, a state-bar auditor, or a malpractice carrier may one day ask you to prove. TrustAI is the substrate you anchor those decisions against. One POST per decision returns a Knox-anchored, ML-DSA-87 signed, Bitcoin-rolled receipt. The audit trail is yours and remains verifiable without depending on TrustAI remaining online.

Knox-anchored decision receipts FRE 902(13)+(14) self-authenticating bundles Multi-vendor cross-validation built in Bitcoin-anchored via OpenTimestamps ML-DSA-87 post-quantum signed Public verify endpoint, zero auth
What you get

An anchor-and-prove substrate beneath your existing agent stack.

TrustAI does not replace your legal-AI agents, your retrieval layer, or your reasoning model. It sits below them. Every substantive decision your agents produce — a clause review, a redline, a risk flag, a cite-check, an attorney override — gets a cryptographic receipt that an examiner can verify independently and that your firm can produce on demand.

Vendor API

POST /api/trustai/v1/decisions/anchor takes one event and returns a receipt with anchor hash, sequence, previous hash, verify URL, and timestamp. Idempotency keys deduplicate retries. Batch up to 100 events per request via POST /api/trustai/v1/decisions/batch-anchor.

Per-matter audit pack

POST /api/trustai/v1/audit-pack/generate bundles every decision under one matter scope into a deterministic zip with manifest, FRE 902(13)+(14) certificate of authenticity, examiner verification instructions, and per-event signed records. Byte-stable: same inputs produce the same bytes.

Public verify endpoint

GET /api/trustai/v1/decisions/verify?hash=<sha256> is zero-auth and returns the full chain block, Merkle inclusion, and Bitcoin OTS proof reference for any anchored payload hash. Opposing counsel can hit it without your help.

Twelve legal-decision event types

A taxonomy of legal-decision event types your agents emit against — clause review, redline, risk flag, cross-validation, attorney override, citation provenance, matter decision, document provenance, privilege scope, conflict check, plus audit-pack lifecycle events.

The audit moat your AI agents are missing

An AI agent without a tamper-evident decision record is a malpractice exposure waiting for a discovery request.

AI-native legal firms are shipping faster than the audit, governance, and state-bar oversight layer underneath them is forming. The first time opposing counsel asks for the model version, the input hash, the reasoning trace, and the cross-validation result behind a contract clause review your agent produced eighteen months ago, the firm that can produce that record and the firm that can't are on opposite sides of a malpractice claim.

Integration

Five steps from your agent stack to a Knox-anchored decision receipt.

  1. Request an API key. Email Jonis directly — there is no procurement queue. You receive a tier-scoped bearer token that authenticates anchor and batch-anchor calls and counts against a rate ceiling.
  2. Wrap your agent's decision-emit path. After your agent finalizes a substantive output, compute a SHA-256 over the input, the output, and the reasoning trace your firm keeps. Pass the digests as opaque tokens — the substrate never receives raw matter content.
  3. POST one event per decision. Send the event_type, the payload (per the event-type schema), and an idempotency key your agent generates. The substrate returns a receipt with anchor_hash, sequence, previous_hash, and a verify_url.
    curl -X POST https://trustailaw.com/api/trustai/v1/decisions/anchor \
      -H "Authorization: Bearer $TRUSTAI_API_KEY" \
      -H "Content-Type: application/json" \
      -d '{
        "event_type": "legal_contract_clause_reviewed",
        "payload": {
          "matter_internal_ref": "<your-opaque-matter-token>",
          "clause_internal_ref": "<your-opaque-clause-token>",
          "model_version": "claude-sonnet-4-6",
          "model_vendor": "Anthropic",
          "input_hash": "<sha256-of-clause-text>",
          "output_hash": "<sha256-of-review-output>",
          "reasoning_trace_hash": "<sha256-of-trace>",
          "risk_score": 0.42,
          "anchored_at": "2026-05-14T21:30:00.000Z"
        },
        "idempotency_key": "<your-unique-key>"
      }'
  4. Store the receipt with your matter record. Your firm's case-management system keeps the receipt next to the decision it covers. The receipt is the production-grade evidence your firm hands over later; the substrate is the verification mechanism that survives independent of you.
  5. Generate a per-matter audit pack on request. When discovery, an audit, or a malpractice query lands, POST the matter scope to /api/trustai/v1/audit-pack/generate and receive a deterministic zip with the FRE 902 certificate, examiner verification instructions, and every signed record under the scope.
    curl -X POST https://trustailaw.com/api/trustai/v1/audit-pack/generate \
      -H "Authorization: Bearer $TRUSTAI_API_KEY" \
      -H "Content-Type: application/json" \
      -d '{
        "matter_internal_ref": "<your-opaque-matter-token>",
        "scope_label": "<short-human-readable-label>",
        "requestor_internal_ref": "<your-opaque-requestor-token>"
      }'

Substrate posture: every *_internal_ref field is shape-enforced to an opaque alphanumeric token. The substrate's schema enforces shape, not content — PII discipline inside the opaque envelope is the firm's responsibility. Customers hash matter, clause, attorney, and requestor identifiers on their side and pass the digest. The substrate anchors what the firm asserts; the firm asserts privilege and identity.

What the examiner, opposing counsel, or state-bar auditor sees

A self-authenticating bundle and a public verify endpoint they can run without your help.

The per-matter audit pack is a deterministic zip with four components, each structured so a reviewer who has never spoken to your firm can verify the record's integrity end to end.

For events whose payload may be surfaced on the no-auth public verify endpoint, the audit pack carries the published payload directly. For matter-internal events — clause review, redline, risk flag, attorney override, citation provenance, matter decision, document provenance, privilege scope, conflict check, cross-validation — the public endpoint returns the anchor and chain integrity without surfacing the payload content. Default-deny on payload preserves matter confidentiality while keeping anchor existence and chain integrity publicly verifiable.

Architectural surfaces

The substrate, in technical terms.

Knox cryptographic anchoring

Per-stream SHA-256 hash chain, hourly Merkle aggregation, Bitcoin anchoring via OpenTimestamps. USPTO provisional application 64/038,359 (no granted patent rights). Independently verifiable at /api/blockchain/verify and /api/blockchain/ots/proofs.

ML-DSA-87 post-quantum signature

NIST FIPS 204 (ML-DSA-87 parameter set) signature on every chain block. Implementation references @noble/post-quantum. Public-key material and verification path are open; signature scheme is post-quantum hardened against Shor's algorithm attacks.

FRE 902(13) and 902(14)

Audit-pack certificate-of-authenticity template tracks the self-authentication framing for records of regularly conducted activity generated by a process or system. The firm's bar-licensed attorney signs; the substrate provides the timing and anchoring evidence.

NIST SP 800-53 Rev 5

Operational posture aligned with NIST SP 800-53 Rev 5 control families relevant to audit-trail integrity, hash chaining, and cryptographic signature. Alignment posture; not certification.

ABA Model Rules 1.1, 1.6, and 5.3

Substrate aligns with competence (1.1), confidentiality (1.6), and responsibilities regarding non-lawyer assistants (5.3 — relevant to AI agents under bar guidance issued in 2024 and 2025). The substrate is the evidence layer; the firm's attorney is the responsibility layer.

Multi-vendor cross-validation engine

Production runs Anthropic Claude claude-sonnet-4-6 primary with OpenAI GPT-5.5 and xAI Grok in parallel as independent cross-validators on every findings-level inference. Cross-checks produce a consensus or divergence anchor publicly readable on the Knox chain. Honest carve-out: vision-extraction perception tasks remain Anthropic-primary because cross-vendor PDF parity is incomplete; downstream findings derived from extracted text run the full multi-vendor cross-check.

USPTO provisional application 64/036,498 (TrustAI), filed 2026-04-11. Inventor of record: Jonis Aaron Fields. Bonis Systems LLC is not yet assignee; assignment is deferred to dedicated IP counsel. Provisional applications do not confer granted patent rights.

Status today

An honest read on what is live, what is open, and what is on roadmap.

No customer-count framing on this page is intentional. The page describes what the substrate is and what it does; what other firms have done with it is not your evaluation criterion.

Pricing on request — this is a working production substrate, not a free trial.