Skip to content

Latest commit

 

History

History
129 lines (99 loc) · 7.8 KB

File metadata and controls

129 lines (99 loc) · 7.8 KB

Roadmap and Project Status

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.

Recent Milestones

  • 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. See docs/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/. The health-records skill can now pull through a self-hosted relay with no third-party dependency, set via HEALTH_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.

Live Skills

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

Self-Hostable Services

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.

Infrastructure

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

In Progress

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

Planned

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

Community Ideas

Component Description
Medication Adherence (IoT) Community proposal
Caregiver Dashboard Community proposal

Community ideas live in Discussions and docs/community-skills.md.

Strategy Artifacts

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

Eval Snapshot: Request-Amendment

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.

See also