1
Inspection
Watch the agents tick in live and trace any signed record back to its provenance.
Inspection →The OS of agents for clinical research
AI-Native CRO Platform
A regulated operating system of agents over commodity GxP systems of record.
Built on 21 CFR Part 11 e-signature gates and ALCOA++ provenance for every regulated action. Agents draft, humans sign, the system remembers — and an inspector can replay any decision.
Live activity
The platform, working in real time
View full timeline →Last 0 provenance events
No data activity
Auto-refreshes every 30sNo activity yet. Wait for the next cron tick.
Architectural invariants
Four rules the platform will never violate
These are not policies — they are enforced in code under packages/policy/ and the orchestrator. A pull request that breaks any of them is rejected by security-review.
01
Agents write only to staging
Promotion to live requires a valid 21 CFR Part 11 signature row. There is no agent path that bypasses the human gate.
02
Provenance for every action
Every regulated write carries created_by, created_at, and provenance_id, and the audit trail is hash-chained and append-only. ALCOA++ is baked into the schema — tamper-evident, not bolted on.
03
Citation grounding is mandatory
Authoring agents flow through packages/citations/. Non-cited claims are rejected before they can be staged.
04
A4 (fully autonomous) is prohibited
Default autonomy class for regulated work is A2 — supervised agentic with a human review and sign gate.
The reference stack
Six layers — L0 through L5
The proprietary moat is L1 through L4: data spine, orchestration, the agent layer, and the accountability fabric. L0 (systems of record) and L5 (experience) are where the platform meets the rest of the world.
- L0Systems of Record
Commodity GxP systems — EDC, CTMS, eTMF, ERP. Integrated, never re-built.
- L1Data Spine
Canonical schemas, vector and graph indices, event bus and outbox. One source of truth across systems.
Moat - L2Orchestration
Inngest workflows, the policy governor, and the staging → live gate. Where work is sequenced and policed.
Moat - L3Agents
31 specialised agents — protocol intake through CSR authoring. Each cited, evaluated, version-pinned.
Moat - L4Accountability
Provenance, audit trail, e-signature, model registry, evaluation. The system that makes the platform inspectable.
Moat - L5Experience
Sponsor portal, operator consoles, inspector view. Where humans review, sign, and decide.
Agent roster
34 agents under one accountability fabric
By autonomy class
Per ADR 0002. A4 is architecturally prohibited for regulated work.
1
A0
Deterministic
5
A1
Assistive
25
A2
Supervised agentic
3
A3
Human-on-the-loop
By cluster
Functional groupings from the AI-Native CRO spec.
- Clinical Monitoring (RBQM)4
- Data Management4
- Medical Writing4
- Start-Up & Regulatory3
- Feasibility & Patient/Site ID3
- Biostatistics3
- Pharmacovigilance & Safety3
- Quality Assurance & Compliance3
- Commercial & Growth3
- Project & Program Management2
- Cross-Cutting Platform2
Build phase progress
Phases 1 → 4
- Phase 14 agents
Foundations
Knowledge, evaluation, audit-trail, essential documents — the platform substrate.
- Phase 212 agents
Core operations
Feasibility, data management, monitoring orchestration, simulation — day-to-day study delivery.
- Phase 310 agents
Clinical and safety
Protocol intake, RBQM, monitoring, PV signal, AE, SUSAR — the regulated clinical surface.
- Phase 48 agents
Authoring at scale
SAP, TLF, protocol authoring, CSR, plain-language summary — citation-grounded medical writing.
Explore the platform
Every major surface, one click away
18 dashboards across overview, operations, sponsor, catalog, and reference. Schematic thumbnails — click any card to open the live page.
Inspection
OverviewEvidence binder an inspector can open cold — every signature, every claim.
Daily Digest
OverviewMorning brief — the last 24 hours of staged work, ranked by what needs you.
Studies Portfolio
OperationsEvery active study with status, health, and per-study consoles.
Live Workspace
OperationsUpload a real protocol — four agents process it at runtime into staged, signed-off work.
Recruitment Funnel
OperationsAgent #34 matches a synthetic, de-identified registry to a protocol and stages tokenized candidates through a consent funnel.
Capture Funnel
OperationsAgents #32/#33 source synthetic opportunities by fit and move them through a human-gated capture & bid funnel.
Activity Timeline
OperationsProvenance events plotted over time — agent by agent, study by study.
Risk Register
OperationsOpen risks, criticals, mitigations — the RBQM surface.
Compliance
ReferencePart 11 and ALCOA+ posture across the staging-to-live gate.
AI Cost Analytics
OperationsToken spend by agent, by study, by model — with cap-burn alerts.
Validation Index
ReferenceCSV/CSA artefacts — per-agent dossiers and evaluation runs.
Architecture
ReferenceSix layers, L0 through L5 — and the moat at L1–L4.
Ops Health
OperationsLive status — services, integrations, cron heartbeats.
Notifications
OverviewUnified feed of every signal worth your attention.
Audit Query
ReferenceAsk the audit trail anything; every answer is a provenance reach.
Sponsor Portal
SponsorSponsor-facing dashboard — requests, deliverables, signatures.
Agent Catalog
CatalogAll 31 agents — autonomy class, cluster, build phase, dossier.
Staging Queue
OperationsAgent-written records awaiting a Part 11 signature.