Build AI systems with evidence built in.
DecisionChain gives AI teams an evidence layer for every production decision — capturing what happened, preserving its integrity, and turning operational records into audit-ready evidence.
Early access · building with design partners. No spam — just product 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
import { DecisionChain } from "@decisionchain/node";
const audit = new DecisionChain({
apiKey: process.env.DECISIONCHAIN_API_KEY,
});
const decision = await model.predict(input);
await audit.log({
model: "risk-scorer@3.2.1",
input,
output: decision,
actor: user.id,
});AI governance breaks when evidence lives outside engineering.
The problem isn't only defining AI governance. It's continuously producing trustworthy evidence that shows how AI systems actually behaved.
- Model decisions happen continuously in production
- Logs are spread across systems and retention windows
- Governance policies are maintained separately from code
- Evidence is collected manually, right before an audit
- Ownership of a given AI decision is often unclear
- Visibility into post-deployment behaviour is limited
The evidence layer for AI systems.
DecisionChain is a drop-in SDK and API-first infrastructure layer that records AI decisions, preserves their integrity, connects operational signals to those decisions, and produces structured evidence for governance and compliance workflows.
Decision logging
Capture the context that matters around each AI decision:
- model / version
- inputs and outputs where appropriate
- actors
- timestamps
- decision metadata
Tamper-evident evidence
Records are append-only and cryptographically chained, so changes to historical records can be detected rather than silently absorbed.
Evidence generation
Turn decision records and governance signals into structured evidence teams can use for audits, internal reviews, customer due-diligence, and compliance workflows.
Governance that stays out of the way.
Add governance to the systems you already build instead of creating another workflow your engineers have to maintain.
SDK / API-first
Everything available through the SDK or the API. No console-driven workflow to maintain.
Minimal integration
Initialise once, record decisions where they happen. No changes to how your models are served.
Runs alongside your stack
Sits next to your existing AI and MLOps infrastructure rather than replacing any part of it.
One evidence layer
Decision records land in a single place instead of being reconstructed from scattered logs.
npm install @decisionchain/nodeimport { DecisionChain } from "@decisionchain/node";
const audit = new DecisionChain({
apiKey: process.env.DECISIONCHAIN_API_KEY,
});
const decision = await model.predict(input);
await audit.log({
model: "risk-scorer@3.2.1",
input,
output: decision,
actor: user.id,
});Four steps from instrumentation to evidence.
- 01
Instrument
Add DecisionChain to the AI application with the SDK or the API.
npm i @decisionchain/node - 02
Record
Capture AI decisions and the relevant context around them as they happen.
audit.log({ model, input, output }) - 03
Verify
Records are cryptographically linked, so unauthorised changes to history are detectable.
append-only · sha-256 chain - 04
Evidence
Use the resulting records as a foundation for governance, audits, and compliance evidence.
structured evidence output
Governance doesn't stop when the model ships.
Most governance work happens after deployment. DecisionChain is built to help teams keep visibility into their AI systems once those systems are live — and we're explicit about what exists today versus what we're still building.
AI decisions
Records of what a model returned, with the context around it.
Model & version changes
Which model version produced which decision, over time.
Historical evidence
Append-only records that stay verifiable after the fact.
Runtime behaviour views
Aggregated views over logged decisions to surface how systems behave in production.
Bias & drift signals
Signal detection across decision records. In development — not a production-grade monitoring system today.
Ownership & accountability
Mapping systems and decisions to the people accountable for them.
Your MLOps stack runs the model. DecisionChain helps prove what happened.
These are complementary layers. Your platform team keeps owning how models are built, shipped, and served — DecisionChain concentrates on the record of what those models decided and whether that record can be trusted later.
MLOps & AI platforms
- Model development
- Deployment and serving
- Performance monitoring
- Infrastructure
DecisionChain
- Decision records
- Evidence integrity
- Governance visibility
- Accountability
- Audit-ready evidence
Built for the teams that build and operate AI systems.
AI / ML Engineering
Integrate governance directly into AI applications without creating another manual process.
Compliance & Governance
Get structured evidence tied to actual AI system behaviour, not to a policy document.
Engineering Leadership
Create clearer accountability and visibility across the AI systems your teams operate.
Our initial focus is AI in regulated environments — healthcare is an early target sector and the EU AI Act is an important entry point — but the evidence layer itself is domain-agnostic.
Built for the evidence requirements behind modern AI governance.
Article 12 of the EU AI Act makes automatic record-keeping over the lifetime of a high-risk AI system an explicit requirement — that logging use case is where DecisionChain starts.
- EU AI ActArticle 12 · logging
- ISO/IEC 42001AI management system
- SOC 2Evidence for controls
- NIST AI RMFGovern · Measure · Manage
DecisionChain is not a certification or a compliance guarantee. It provides infrastructure and evidence that can support your existing governance and compliance processes.
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.
Read post → - ·5 min read
Hash Chains 101: How Immutable Logs Actually Work
You don't need a blockchain. You need a hash chain. Here is the mechanic that makes an audit log verifiable — in one page.
Read post → - ·6 min read
What to Log for Every AI Decision
The record shape auditors actually want, and the fields that will save you when a decision is challenged months later.
Read post →
Testing decision reconstruction for high-impact AI workflows
We are working with a small number of AI teams to test whether a tamper-evident decision record can capture the AI recommendation, human review, final action, and context needed for later governance review.
Initial evaluations use synthetic, redacted, staging, or customer-controlled data.
No spam. Just launch updates.