Developer notes / Updated 7 October 2026

Clear boundaries.
Inspectable systems.

The architecture behind our software, the role Claude plays today and the model integrations we plan to build next.

Public transit,
native delivery.

Current app architecture

Departure times and service notices come from provider APIs. A language model is not the source of live departure information.

Source Transport provider APIs

Regional schedules, departures and notices

Application logic Normalize & validate

Source labels, timestamps and availability states

Native delivery iPhone / Widgets / Live Activities

Readable departures in the places people use

Regional sources.

Public transport portfolio and source boundaries
Edition Data source Product scope
Stockholm SL transport integration API Departures, favourites, widgets and Live Activity
Finland Regional transport feeds and Finnish rail data Regional departures, nationwide trains and service notices
Norway Entur Departures and supported ferry services; in development
Vienna Wiener Linien Open Data Departures and separate lift/escalator disruption information; in development
Hong Kong Operator feeds and Transport Department notices Supported bus, rail, minibus and ferry arrivals; in development
LondonTfL Open DataBus, Tube, DLR, Elizabeth line and London Overground departures; in development

Independent apps. Provider coverage and availability vary; the apps do not sell transit tickets.

Building
with Claude.

Current use

Claude supports software engineering, story planning and production tooling. OpenFilmQA can be installed as a Claude skill and provides local evidence packets, structured observations, repair comparison and review coverage.

Planned API integration

Fate Council is our primary next Claude API product. Deterministic engines compute chart symbols; a server-side Claude Sonnet integration is planned to turn selected symbols and a user’s question into contextual reflection. Model interpretation stays separate from factual claims.

Medmate targets Claude API chat for app help in its next update. Trúc Lâm already has AI question answering; integrating Claude into that experience is the next development goal.

OpenFilmQA has a separate Claude API research track: media evidence and director intent become structured findings. Validation rejects malformed results and missing coverage; the filmmaker retains the release decision.

A future transit experiment would parse public agency bulletins into structured disruption fields using Claude. This is research, not a deployed feature, route planner or prediction service.

Explore the Claude review workflow

Questions inside
every app.

Planned Claude API development

Our portfolio-wide AI API direction is a question-answering layer tailored to each app: transit notices, Trúc Lâm library passages, calculation explanations, Medmate routines, Fate Council reflection and film production evidence.

Trúc Lâm’s existing AI chat is available today. The planned Claude API integration is a separate upgrade; the current model and data path are not represented here as Claude.

Proposed request path

User question and chosen context → authenticated application service → server-held Claude API key → Claude response → validation, source links and an answer in the app.

The key will stay on the server. The model receives only the context the user selects for the question. User permission, rate limits, data handling and usage costs must be defined before release.

Transit answers will cite provider notices. Trúc Lâm answers will point back to library sources. Numerical answers will be checked by the calculation engine. Meds Reminder’s scope will be app usage and routines, not clinical guidance.

This is a shared development goal. It is not a statement that every released app already calls Claude.

Privacy at the
product boundary.

Meds Reminder stores medicine records locally and recognizes prescription text with Apple Vision on device. Today, it does not send medical queries to Claude. The planned next-update chat covers app settings and reminder workflows, not clinical questions or dose recommendations.

OpenFilmQA runs locally. Connecting an external reviewer is an explicit action using the maker’s own authorized account, with the reviewer’s data handling applying to that connection.

Each app’s own privacy policy describes its data flows. We do not make a blanket zero-retention promise across providers or services.

Read the website privacy notice