Prove every AI decision.
Down to the byte.
Constable is a tamper-proof cryptographic log that tracks every execution of your AI models — a monotonic, replayable history of usage that turns compliance from a fire-drill into a query.
- Monotonic
- append-only
- Replayable
- full history
- Verifiable
- merkle-rooted
Built for compliance teams in the most regulated industries
The accountability gap
Your AI makes thousands of decisions a minute. Can you prove what it did?
Regulators no longer accept “the model handled it.” Application logs are mutable, dashboards are point-in-time, and GRC checklists prove a control exists — not what actually happened. When an examiner asks for a defensible, unaltered history, most teams have nothing.
Mutable logs
Standard logs can be edited, dropped, or backfilled — so they can't stand as evidence.
No ordered truth
Without monotonic ordering, you can't prove the sequence of what your models did.
Audit scramble
Every review becomes a manual scavenger hunt across teams, tools, and screenshots.
The platform
One ledger of record for every model execution
Constable captures a cryptographically-chained record of what your AI did — and proves it hasn't changed. Not another dashboard: the evidence layer underneath them all.
Monotonic hash chain
Every execution is sealed to the one before it with a strictly increasing height. Reordering, backdating, or deleting an entry breaks the chain — and is detected instantly.
Deterministic replay
Reconstruct the complete, ordered history of any model, service, or time window on demand. Hand an examiner a reproducible timeline instead of a screenshot.
Tamper-evident by design
Merkle-rooted, append-only storage means the log proves its own integrity. If a single byte changes, the root no longer verifies.
Model-level provenance
Capture the model, version, actor, prompt fingerprint, and outcome for each call — the evidence GRC checklists and SIEM dashboards never hold.
Attestation templates
Turn ledger evidence into examiner-ready language with out-of-the-box templates mapped to the controls each framework requires.
Zero-trust ingestion
A lightweight SDK wraps existing model calls. Keys stay server-side, entries are signed at the edge, and nothing sensitive is exposed to the client.
Pre-built compliance playbooks
Turnkey audit configs for the frameworks that matter
Ship day one with out-of-the-box attestation templates per vertical. Each playbook maps ledger evidence to the exact controls your examiners ask about.
HIPAA
Prove PHI-handling AI stays inside policy with access, minimum-necessary, and audit-control evidence.
- §164.312 audit controls
- PHI access attestations
- Breach-timeline replay
SOX
Demonstrate change-integrity and segregation of duties for AI in financial reporting workflows.
- ITGC evidence pack
- Controller sign-off trail
- Close-cycle replay
FINRA
Supervise AI touching client communications with suitability and books-and-records coverage.
- Rule 3110 supervision
- Comms review ledger
- 17a-4 retention
LPP
Defend privilege with a sealed record of every AI decision over discovery and privileged material.
- Privilege redaction log
- Chain-of-custody proof
- Reviewer attestations
Where Constable fits
GRC checks the boxes. SIEM searches the noise. Constable proves the truth.
Constable complements the tools you already run. It fills the gap they leave open: per-execution, tamper-evident AI evidence.
* GRC automation, e.g. Vanta & Drata. ** SIEM / log analytics, e.g. Splunk. Comparison is illustrative and reflects category focus, not any single product's full feature set.
Attestation Copilot
From ledger evidence to examiner-ready language
Generative AI drafts the attestation; the cryptographic ledger backs it up. Try it live below — it runs on your own OpenAI account, so nothing is ever fabricated.
Draft an audit-ready attestation
Pick a framework, describe your AI system, and Constable drafts defensible attestation language referencing the cryptographic execution ledger.
Fictional sample data. Drafts are AI-generated starting points, not legal advice.
Your generated attestation will appear here, ready to paste into your evidence binder.
Social proof
Trusted by the people who sign the attestations
“Our examiners stopped asking for screenshots. We hand them a replayable ledger and the conversation is over in minutes.”
“We could finally answer 'what did the model do, and can you prove it wasn't changed?' with math instead of a policy PDF.”
“The HIPAA playbook shaved weeks off audit prep. The evidence was already mapped to the controls our auditor cares about.”
Pricing
Usage-based. Aligned to your AI footprint.
One simple model for regulated enterprises. Pay for the executions you record — nothing you don't.
+ $2.50 per million API calls logged
- Tamper-proof, monotonic execution ledger
- Deterministic replay & Merkle verification
- HIPAA, SOX, FINRA & LPP playbooks included
- Attestation Copilot with your own AI keys
- SDKs, SSO, and role-based access
- Dedicated compliance success manager
Scales from pilot to enterprise
Teams start around a few million calls a month and grow into billions as AI moves into production. Volume and committed-use discounts available for large deployments.
Talk to us for annual and committed-use pricing.
FAQ
Questions, answered
Vanta and Drata automate control checklists and evidence collection for org-level certifications. Constable operates one layer deeper: it captures a cryptographically-chained, tamper-evident record of every individual AI model execution. Where GRC tools tell an auditor a control exists, Constable proves — mathematically — exactly what your models did, in what order, and that the history has not been altered.
Splunk indexes logs for search and observability, but application logs are mutable and can be edited, deleted, or backfilled. Constable's ledger is append-only and monotonic: each entry seals the one before it, so any tampering breaks the chain and is instantly detectable. It's built for evidence, not just search.
Every execution is assigned a strictly increasing height and linked to the prior entry's hash. Nothing can be inserted, reordered, or removed after the fact. Replay lets an auditor reconstruct the complete, ordered history of usage for any model, service, or time window on demand.
Constable ships turnkey audit configurations and attestation templates for HIPAA, SOX, FINRA, and LPP out of the box. Each playbook maps ledger evidence to the specific controls examiners ask about, so your compliance team starts from a working baseline instead of a blank page.
Usage-based: a $2,500/month base tier plus $2.50 per million API calls logged. You only pay for the executions you actually record, which keeps costs aligned with your AI footprint as it scales.
Most teams emit their first sealed ledger entries in an afternoon using our SDKs and a lightweight middleware wrapper around existing model calls. No changes to your model providers are required.
Make your AI provable.
Join compliance officers, data scientists, and IT leaders getting ahead of AI regulation. Reserve your spot and we'll tailor a walkthrough to your framework.