HELIX
Now Available · v1.0

Behavioral Fingerprinting For Production AI.

Helix instruments your existing AI pipeline and produces cryptographically verifiable behavioral archives. No model changes. No silent failures. Know when your system changed. Prove it.

Try Open Source Speak to us about Enterprise →
145
Days until EU AI Act
enforcement · Aug 2, 2026
6 mo
Minimum log retention
required under §12
€15M
Maximum penalty for logging
non-compliance (Art. 71)
<5 min
From pip install
to first signed archive
helix-trial ● live

How Helix Works

A complete audit trail,
from first span
to final proof.

Helix connects to your existing observability stack and outputs a tamper-evident archive with behavioral fingerprints your auditors can verify, without access to Helix infrastructure, without a vendor call.

01

Capture

Helix normalizes span trees from any AI telemetry source. No schema changes on the emitting system.

OTLP · gRPC Langfuse JSONL No SDK
02

Fingerprint

Each span tree is reduced to an immutable template: structural shape, separated from volatile runtime values.

SemanticTraceCodec Jaccard similarity Drift alerts
03

Archive

Every run is sealed into a signed bundle with operator ID, timestamp, and retention floor locked at write time.

.eplbundle SHA-256 6-month floor
04

Prove

Any archived bundle can be reopened and verified offline. The fingerprint is re-derived and compared to the stored hash.

Offline replay Auditor-portable No dependency

Pre-Production Validation

Tested on real production
trace corpora.

These numbers come from validation runs against three real-world trace formats. Not synthetic benchmarks, not curated samples. Helix was tested on production data before any enterprise deployment.

OTLP Fidelity
91.6%
OpenTelemetry production corpus. Behavioral fingerprint accuracy across full span tree reconstruction.
Langfuse Accuracy
91.4%
Langfuse / LangGraph trace export. Template extraction across multi-step agent workflows.
Templates Extracted
24
Distinct behavioral templates from a 1,847-span corpus across 312 individual production traces.
Time to First Archive
<5min
From pip install to first signed .eplbundle on a new machine with existing traces.

EU AI Act · Article 12

Enforcement begins
August 2, 2026.
That is not an estimate.

High-risk AI systems must maintain automatically generated, verifiable logs for a minimum of six months. Non-compliance is a regulatory event, not a technical inconvenience.

145
days until enforcement
August 2, 2026 · EU AI Act Article 12
Operators must produce verifiable logs demonstrating behavioral consistency, chain of custody, and a minimum six-month retention period, on demand, to a national supervisory authority.
Article 12 Requirement Helix
Automatic logging of AI system events
§12(1)(a) · no manual intervention permitted
 Implemented
Minimum 6-month log retention
retention_floor_date locked at write time
 Implemented
Verifiable integrity of logs
SHA-256 canonical hash, tamper-evident .eplbundle
 Implemented
Operator identification in records
operator_id embedded per archive at write time
 Implemented
Behavioral consistency monitoring
Jaccard drift detection · 0.95 compliance threshold
 Implemented
Offline verification without vendor
Bundle replay with zero Helix infrastructure dependency
 Implemented

Editions

Start open. Scale with confidence.

Open Source · Apache 2.0
Helix Trial

Run a real ingest and archive test against your own traces. Confirm Helix reads your format, generates a verifiable bundle, and returns a clear workload-fit verdict in under five minutes.

  • Format detection: Langfuse JSONL, OTLP JSON, OpenInference, flat JSONL
  • .eplbundle archive generation and offline replay verification
  • Workload-fit verdict: pilot_now / narrow_pilot / not_fit_yet
  • HTML summary report: human-readable ingest and archive results
  • Docker support: single command, no local dependencies
View on GitHub
Enterprise
Helix Enterprise

Production-grade deployment with SLA, cloud storage integrations, real-time trace routing, and compliance reporting.

  • SemanticTraceCodec: full behavioral fingerprinting pipeline
  • Drift detection: configurable Jaccard thresholds, real-time alerts
  • Helix Gateway: live OTLP routing, S3 / GCS / Azure Blob archive
  • EU AI Act Article 12 compliance mode & report export
  • Dedicated support, SLA, and onboarding
Request Enterprise Access

Make your AI
system auditable
before August.

The open-source trial takes an afternoon. Enterprise onboarding takes a week. The compliance deadline is not negotiable.

Start with Open Source Speak to us about Enterprise
helix@traceintegrity.org  ·  GitHub