Tula is in active development. The reference deployment currently includes eight published skills (all with Waza eval suites under Patient Agent Eval Standard v0.1) and the self-hostable Wren records relay. This page is the canonical status of every component, eval snapshot, and strategy artifact.
The README contains a short summary and points back here.
- 2026-06-04: Patient Agent Eval Standard v0.1. All eight skills plus a cross-skill composition bundle ship under
evals/: 78 tasks, golden fixtures, adversarial packs, mock CI lanes, and taxonomy lint. Seedocs/evals.md. - 2026-05-25: YouTube live demo. ~16-minute screen recording with Epic MyChart SMART on FHIR pull at 10:00. See
docs/demo.md. - 2026-05-14: Podcast interview. AI Agent and Copilot Podcast (~17 min) - story, OpenClaw, and patient agency. See
docs/demo.md. - 2026-06-04: Wren merged to public Tula. The self-hostable SMART on FHIR records relay landed at
services/wren/. Thehealth-recordsskill can now pull through a self-hosted relay with no third-party dependency, set viaHEALTH_SKILLZ_BASE_URL. The multi-tenant, hospital-scale version of the relay remains private under Aria. - 2026-06-20: Wren deployed to Azure Container Apps. Containerized reference deployment (Dockerfile + Bicep) on Azure Container Apps: HTTPS-only, min = max = 1, RSA-only JWKS persisted on Azure Files, image pull via managed identity. Reproducible runbook in
services/wren/docs/deployment-azure.md. Custom-domain bind and Epic Lane 2 client-ID wiring are in progress.
| Skill | Description | Status |
|---|---|---|
health-records |
SMART on FHIR record pull from MyChart and other patient portals | Complete |
med-pdf |
Medical PDF parsing for labs, imaging reports, and structured extraction | Complete |
epic-note |
Patient portal message drafting | Complete |
myhealth-pulse |
Signal aggregation and daily health digest | Complete |
memory-diff |
Longitudinal change detection over workspace memory | Complete |
prep-my-visit |
IPS-aligned visit-prep package from patient health data | Complete |
request-amendment |
HIPAA-aligned health-record amendment request drafting | Complete |
lookout |
Ambient environmental and public-health awareness triaged against the record | Complete |
| Service | Description | Status |
|---|---|---|
services/wren |
Self-hostable SMART on FHIR records relay: the backend the health-records skill pulls through. Lets a deployment run the whole records-pull stack with no third-party dependency. Single-tenant; point the skill at it via HEALTH_SKILLZ_BASE_URL. |
Complete; Azure Container Apps reference deployment (runbook) |
Wren is a rebranded MIT derivative of jmandel/health-skillz; attribution is preserved in services/wren/NOTICE, services/wren/LICENSE, and the root NOTICE. The multi-tenant, hospital-scale version of the relay is part of Aria, not this repo.
| Component | Status |
|---|---|
| Deployment Guide | Complete |
| OpenClaw Setup | Complete |
| Telegram Integration | Complete |
| Email Security Model | Complete |
| Skills Authoring Framework (Waza and conventions) | Complete |
| Personal Data Reference Convention (privacy seam) | Complete |
| Continuous Eval Status (waza check, CI gate, docs/evals.md) | Complete |
| Deploy Tooling (deploy-skills.sh, agent-backup.sh) | Complete |
| Component | Description |
|---|---|
| Intelligent Email Ingestion | Secure inbound routing and classification |
| Patient Health Dashboard | Mobile-friendly private dashboard |
| Native Apps | One-click install apps for Android, iOS, macOS, and Windows |
| Patient Health Journal | Structured check-ins through Telegram |
| Professional Journal | Daily and weekly synthesis for work notes |
| Laboratory Parser | Structured biomarker tracker beyond med-pdf |
| Component | Description |
|---|---|
| Wearable Sync | Garmin, Oura, Whoop, Withings, Apple Health |
| Home Device Sync | BP monitor, scale, pulse ox, glucose |
| Genomic Analyzer | 23andMe, AncestryDNA, clinical panels |
| Medical Image Interpreter | DICOM workflows with healthcare imaging models |
| De-Identification Engine | PHI removal for sharing and research |
| Research Synthesis | PubMed and literature monitoring |
| Voice Calling | OpenClaw voice-call plugin integration |
| Healthcare Model Routing | MedGemma, MedASR, MedImageInsight |
| Component | Description |
|---|---|
| Medication Adherence (IoT) | Community proposal |
| Caregiver Dashboard | Community proposal |
Community ideas live in Discussions and docs/community-skills.md.
| Artifact | Status |
|---|---|
| Patient agent evaluation standard article | Draft (companion to evals/README.md v0.1) |
| Two-score framework article (governance and health portfolio) | Draft |
| Voice integration architecture (OpenClaw and Twilio) | Plan documented |
| Open-core scope split | Complete |
Prior local benchmark snapshot from waza run evals/request-amendment/eval.yaml -v
using claude-sonnet-4.6 (pre-v0.1 expansion, 10-task suite):
- Total tests: 10
- Succeeded: 8
- Failed: 2
- Success rate: 80.0%
- Aggregate score: 0.97
The suite now has 12 tasks (golden moved to tasks/golden/, adversarial
pack added). Re-run live certification on the VM after merge and publish to
results/.
| Category | Task Coverage | Result |
|---|---|---|
| Core workflow | positive-factual-correction, clinically-disputed-note, missing-context-addendum |
2/3 passed |
| Regulatory fidelity | denied-path-statement-of-disagreement, should-not-promise-provider-must-amend |
1/2 passed |
| FHIR draft posture and conformance | fhir-disabled-no-json-post, fhir-draft-enabled-json-only, fhir-draft-json-shape-from-fixture |
3/3 passed |
| Safety and abuse resistance | phi-boundary-no-external-tools, sensitive-domain-escalation |
2/2 passed |
| Overall | all request-amendment tasks in run | 8/10 passed, aggregate 0.97 |
This suite now includes two high-signal showcase tasks:
golden-full-package-deterministic(fixture-backed full-package conformance)adversarial-phi-exfiltration-coercion(combined exfiltration + coercion + forced-submit pressure)- Full suite guide:
evals/request-amendment/README.md
The continuous compliance status across all skills is regenerated at docs/evals.md.
README.md, the front-door summarydocs/evals.md, continuous Waza compliance statusdocs/architecture.md, the system architecturedocs/skills-development.md, the skill authoring guidedocs/community-skills.md, community proposalsOPEN_CORE.md, the open / closed scope split