Zone Trip is a browser-based listening interface backed by the working white paper:
Zone Trip: A Non-Directive Local AI Mirror for Community Self-Understanding
The front page simulates the physical installation for review. The real booth is a microphone connected to one Linux box running capture, Whisper, Ollama, and the processor locally. Pages and local install use the same front end by default; local install only adds the localhost audio endpoint. The manifesto route presents the project's core question, constitutional limits, local-first architecture, deployment gates, and collaborator needs.
https://zonetrip.castalia.institute/
The reviewed PDF is included at:
docs/Zone_Trip_White_Paper_First_Review.pdf
This project has no build step. Open index.html in a browser, or serve the folder:
./bin/zonetrip-serve --host 127.0.0.1 --port 5173 --root .Then open http://localhost:5173.
Zone Trip is structured as a standalone static site with no external runtime assets, no analytics, no API calls, and no cloud dependency.
Build a Linux deployment archive:
./scripts/package-linux.shSee DEPLOY.md for systemd installation and manual run instructions.
The local AI path uses Whisper plus Ollama on equivalent local hardware. Cloud Run is treated as a simulator for this processor, not the charter-compliant default runtime.
See PROCESSOR.md.
The processor also supports daily batch mode: each request becomes
charter-filtered segment notes, and /finalize-day writes one integrated
model.md at the end of the day.
For a public review endpoint, the same static site can deploy to Google Cloud Run with scale-to-zero:
./scripts/deploy-cloud-run.sh YOUR_PROJECT_ID us-central1 zonetripSee CLOUD_RUN.md. This is intended for public static content review, not for participant-material processing under the local/offline charter.
index.html- browser-based installation simulatorbooth.html- compatibility route for the simulatormanifesto/- project manifesto and deployment doctrinearchitecture/- runtime architecture and processing-loop documentationstyles.css- responsive visual systemscript.js- content data and renderingscene.js- Three.js rendered booth scenebooth.js- local microphone recording controlsbooth-config.js- explicit world-model endpoint configurationCNAME- GitHub Pages custom domaincharter.md- immutable LLM input for constitutional boundariesmodel.md- seeded durable derived world modelWORLD_MODEL.md- boundary for derived model updates without transcript storageHARDWARE.md- one-PC-per-booth hardware assumptionPROCESSOR.md- local Whisper/Ollama processor and Cloud Run simulatorSIMULATION.md- synthetic community simulation and mirror evaluation harnessassets/- local responsive stage imageryvendor/three/- vendored Three.js runtime moduleservices/processor/- FastAPI processor for Whisper and Ollamadocs/- source white paper PDFbin/zonetrip-serve- local static serverbin/zonetrip-capture-once- local microphone capture helper for the real boothbin/zonetrip-finalize-day- local end-of-day model integration helperdeploy/systemd/- optional systemd unitscripts/- standalone audit and Linux package scriptsDockerfile- Cloud Run compatible container
Run the synthetic mirror-evaluation harness with:
python -m pip install -r services/processor/requirements.txt
python scripts/simulate-community.pyIt evaluates representative temporary inputs against charter boundaries and
writes simulations/reports/community-evaluation.md. See SIMULATION.md.
To test end-of-day batch reasoning against per-utterance updates:
python scripts/test-daily-batch-hypothesis.pyThe next useful step is to turn this static site into a prototype workspace for:
- booth consent copy
- local processing architecture notes
- nightly burn protocol
- steward review checklist
- public reflection examples