Immutable audit trails for AI decisions.
Prove your models are auditable in 30 minutes.
DecisionChain gives AI teams tamper-evident logs of every model decision — with bias detection and one-click evidence for the EU AI Act, ISO 42001, and SOC 2. Drop in the Python SDK, ship your models, hand auditors a link.
Early access: 50% off your first 6 months
View live demo →Design partners get founder-level support. No spam — just launch updates.
- dec_01HR9F4Q…10:14:22.481risk-scorer@3.2.1review · 0.82a91f…c204
- dec_01HR9F4P…10:14:22.418risk-scorer@3.2.1approve · 0.11b03e…7d1a
- dec_01HR9F4N…10:14:22.302risk-scorer@3.2.1approve · 0.24c8f2…9e40
- dec_01HR9F4M…10:14:22.198risk-scorer@3.2.1review · 0.711d77…22ab
from decisionchain import DecisionChain
audit = DecisionChain(api_key=os.getenv("DC_KEY"))
decision = model.predict(input)
audit.log(
model="risk-scorer@3.2.1",
input=input,
output=decision,
actor=user.id,
)Compliance infrastructure that lives next to your model, not next to your policies.
DecisionChain is a drop-in SDK and hosted service that records every decision your models make into an append-only, cryptographically chained log. It runs bias and drift checks in the background, and turns those records into audit-ready evidence for the EU AI Act, ISO 42001, and SOC 2 — without your team building any of it in-house.
Immutable by design
Every record is hashed and chained to the previous one. Tampering breaks the chain and every downstream signature.
Bias & drift, watched
Continuous checks across your logged decisions, with thresholds and alerts you can point an auditor at.
Evidence on demand
Export a signed evidence pack scoped to a model, a customer, or a date range. Ready for regulators or buyers.
Four steps from first log to audit-ready evidence.
- 01
Install the SDK
One pip install, one API key. Python-first, with a REST API for anything else.
pip install decisionchain - 02
Log every decision
Wrap your model call. Inputs, outputs, versions, and actors are captured with a single log() call.
audit.log(model=..., input=..., output=...) - 03
Tamper-evident storage
Records are hashed, chained, and anchored to an external witness. Alteration breaks the chain, provably.
sha-256 chain · anchored hourly - 04
Export compliance evidence
Generate signed evidence packs for the EU AI Act, ISO 42001, and SOC 2. Share a link, not a spreadsheet.
audit.evidence.export(framework='eu_ai_act')
Evidence for the frameworks your buyers ask about
- EU AI ActArticle 12 · logging
- ISO/IEC 42001AI management system
- SOC 2Type II · security
- NIST AI RMFGovern · Measure · Manage
Framework references, not certifications. DecisionChain produces evidence your compliance team submits.
Notes on AI compliance and audit infrastructure.
- ·7 min read
Why Tamper-Evident Audit Trails Matter for AI Compliance
Mutable logs are stories. Chained, anchored records are evidence. Here is what tamper-evidence actually means for AI systems — and why it is quickly becoming the default requirement.
Read post → - ·8 min read
What the EU AI Act Actually Requires From Your Logging Infrastructure
Article 12 says 'automatic recording of events.' Regulators are already telling providers what that means in practice. Here is the plain-English version for engineers shipping high-risk systems.
Read post → - ·6 min read
ISO 42001 vs SOC 2: What AI Companies Need to Know
You already have SOC 2. Your enterprise buyers are starting to ask about ISO 42001. Here is what overlaps, what is new, and where AI companies get caught off guard.
Read post →
Ship auditable AI before your first compliance conversation.
We're onboarding a small group of AI teams for early access. Design partners get founder-level support and 50% off the first 6 months. Drop your work email and we'll be in touch.
No spam. Just launch updates.