Protocol to submission, without the middleware.
CDISC-native agents take a protocol to SDTM, ADaM, and Define-XML in under two weeks — not 68 days. Every artifact carries a Part 11 audit trail your regulatory reviewer can follow without a guide.
Three products. One data layer. One audit trail.
Core writes every artifact a trial produces. Signal reads across every function that runs it. Agent Studio builds the workflows specific to yours. One data layer, one audit trail, across all three.
Core
Seven specialized agents. Protocol → USDM → DMP → eCRF → SDTM → ADaM → Define-XML. Study builds in under two weeks. Every artifact carries its reasoning trail.
Signal
One data layer, six role-aware views. Every function opens Signal and finds the narrative already written. The Monday-morning briefing in 60 seconds, not two days.
Agent Studio
The agents your trial needs that Core doesn't cover. Build them in the same Part 11 runtime, with the same Decision Queue and the same audit trail. Your workflows. Your agents.
Also on the OS: the iOS Decision Queue. Built by the TrialNexus team. Book a protocol walkthrough.
The people running your trials are exceptional. The tools they're using are not.
Clinical-trial teams are some of the most rigorously trained professionals in medicine. They spend most of their day reconciling spreadsheets.
These are not abstract inefficiencies. A 68-day study build is a 68-day delay before your trial can generate safety data. An 85% miss rate on enrollment timelines means most trials cost more and take longer than planned. TrialNexus closes these gaps without asking your team to work harder.
Seven things that make it CDISC-native, not ML-bolted-on.
The whole platform at a glance — every capability traceable to a real artifact your team and your regulator can follow.
Narratives, not dashboards
Every view in Signal comes pre-read. Weekly briefings per study, ad-hoc answers to questions, citations back to the SDTM variable. The Monday-morning briefing in 60 seconds.
CDISC-native architecture
USDM-centric. CDASH, SDTM, ADaM and Define-XML are first-class citizens, not afterthoughts bolted onto a general ML stack. Define-XML that passes Pinnacle 21 on the first run.
Audit trails built in
Every decision logged with provenance and reasoning. 21 CFR Part 11 and ICH E6(R3) compliant by design. An audit trail your regulatory reviewer can follow without a guide.
Human-agency control plane
Manual · supervised · autonomous. You set the boundaries per agent, per study. Context windows are scoped so agents only see what their task requires.
Enterprise integrations
Deploy into Medidata Rave (ODM XML) and Veeva Vault CDMS. Auth0 SSO. Query cycles that close in hours, not days.
Explainable by default
Every artifact carries a reasoning trail. Agents show their work: regulators see every step. Every decision your DM lead makes comes with the reasoning already written.
Open, not locked-in
Your data stays in your systems. USDM makes protocols portable. If you leave, you leave with everything. Every artifact in open formats, no extraction fee.
You are in control. The agents work within the boundaries you define.
TrialNexus is built on a four-layer control architecture: CDISC-standards compliance enforced at the model level, context-window scoping so agents only see what they need, a human-approval gate at every decision point, and an immutable audit trail that shows regulators exactly what was automated and what was reviewed.
drag the lever ←→ or tap a label
Routine actions auto-execute above your configured confidence threshold. Edge cases route to the Decision Queue. The default for most active studies: agents handle the routine they can do reliably, your team owns the decisions that require expertise.
Most AI tools optimize for autonomy. We optimize for defensibility. In a regulated trial, the question is never "can the agent do this?" — it's "can you show the FDA exactly how this decision was made?" Our control plane answers that question before anyone asks.
Every value traces back to the exact protocol sentence.
When an inspector asks "how was this derived?", you don't reconstruct it. You open the graph.
Drop into your stack. Leave it standing.
TrialNexus is additive. We meet your EDC where it lives and keep the data where you keep it.
@TrialNexus from the tools your team already uses.
Every agent available as an MCP endpoint. Invoke complex clinical workflows from Claude Enterprise, ChatGPT, or any MCP client — with full CDISC validation and an unbroken audit trail.
Works where you already work
Invoke @TrialNexus from Claude Enterprise, ChatGPT, or any MCP-compatible client. No new interface. Your team stays in the tools they know.
Tasks base models cannot complete
SDTM mapping, SoA extraction, EDC population — these require CDISC domain knowledge and multi-step workflow logic no frontier model carries natively.
Audit trail on every call
Every invocation logged at the token level: model version, inputs, outputs, agent version, human approval. The same immutable trail as the full Core pipeline.
Writing from the translation layer.
CDISC, Part 11, and the engineering of trustworthy agents — no gated PDFs.
Your enterprise AI is writing emails. It should be closing study builds.
Sponsors pay five figures a year for frontier-model access and point it at meeting notes. Nexus Assistant makes every CDISC-native TrialNexus agent callable from Claude, ChatGPT, or Gemini — with a Part 11 signature behind every approval.
Is your AI clinical trial tool actually 21 CFR Part 11 compliant? A checklist.
Most AI vendors claim Part 11 compliance. Here is what the regulation actually requires for AI-generated clinical data artifacts — and the questions to ask before you sign.
The SDTM submission errors that cause Pinnacle 21 failures — and how to prevent them
A breakdown of the most common SDTM conformance failures, why they happen, and what a prevention-first pipeline looks like.
What ICH E6(R3) actually requires for AI-assisted clinical data management
The GCP guideline finalized in January 2025 has specific implications for sponsors using AI tools in their data management pipeline.
Agentic AI in clinical trials: what it actually means for your data management team
A precise definition of what agentic AI can and cannot do in a regulated trial, and how to evaluate a platform that claims it.
Why human-in-the-loop is the only viable model for AI in regulated clinical trials
Autonomy and compliance are not opposites — but getting both requires an architecture most AI tools have not built.
Why USDM v4.0 is the most important standard in clinical trials
A walkthrough of the Unified Study Definitions Model and what changes when protocols are machine-readable by default.
Building an audit trail you would actually show a regulator
How we designed immutable provenance records on top of Postgres, and the tradeoffs we made for Part 11.
The agency dial: how to give agents responsibility without losing oversight
The user-interface primitive we keep coming back to, and why it matters more than any model benchmark.
What goes wrong in a 68-day study build
Anonymized cycle-time data from conversations with DM leads at six mid-size sponsors.
SDTM mapping is pattern-matching with consequences
Why LLM-assisted SDTM mapping is hard, where it works, and where it should never be autonomous.
Introducing TrialNexus — people were never meant to be middleware
Our founding post. Why the next ten years of clinical data management belong to agent-based pipelines.