Project Info
Inspiration
At least once in our lives, each of us has felt alone and lost. Maybe it lasted a minute. Maybe a night. Maybe much longer. It is that moment when the room suddenly feels too large, the phone too quiet, and even the smallest decision weighs somewhere between the stomach and the chest. I started from a simple, perhaps stubborn idea: we may not be able to prevent every difficult situation, but feeling lost should never mean being left alone. That is where NotAlone was born. I also believe that doing good travels in both directions. It helps the person who receives it, of course, but it changes something inside the person who offers it too. A gesture, a little time, the right words at the right moment: small things, perhaps, but small does not mean insignificant. NotAlone is my attempt to give technology this purpose. Not to replace people. To help people be there.
What it does
NotAlone listens before it organizes. The person can describe a difficult situation naturally: being stranded abroad, losing documents, facing a family problem, feeling overwhelmed or simply not knowing where to begin. Before creating a plan, NotAlone builds a Human Agency Contract. It understands the person’s current capacity, preferred support style, consent and privacy boundaries. It then creates the smallest useful team of specialized AI agents for that exact situation. The agents examine different parts of the problem, challenge uncertain information and distinguish facts from assumptions. But the person does not receive another heroic forty-seven-step checklist. When your head is already burning, step thirty-eight might as well be on the Moon. NotAlone begins with one reversible, manageable action: the Next Kindest Step. If the person says, “This is too much,” the Living Plan becomes smaller and adapts without forgetting the boundaries that were already expressed. No automatic emergency call, third-party contact or location sharing happens without explicit human approval.
How we built it
The experience is built with Flutter, with a FastAPI, Python and Pydantic backend. Codex and GPT-5.6 accompanied the project throughout Build Week as an engineering collaborator. They helped inspect the inherited foundation, separate the new Build Week scope, design the human-agency and situation models, implement the dynamic agent runtime, strengthen retries and fallback behavior, create focused tests, verify the Vercel deployment and review the claims shown to judges. The human decisions remained human: the purpose of the product, its name, its boundaries, its consent model, its privacy rules and the actions it must never take by itself. The hosted runtime is honestly attributed. It currently uses trace-confirmed openai/gpt-oss-120b through Groq or a clearly labeled deterministic fallback. It never presents GPT-OSS as GPT-5.6. Challenges The hardest part was not making the AI do more. It was teaching the product when to do less. A person in difficulty does not necessarily need the most complete answer. They need an answer they can actually carry, without being frightened, overwhelmed or pushed beyond their consent. We also needed to make the AI team transparent. NotAlone therefore shows how the agents worked, which execution mode was used, what remains uncertain and where fallback occurred. Accomplishments During Build Week we completed: a working public Flutter and FastAPI application; the Human Agency Contract; situation-specific agents and subagents; independent execution traces; a critic and uncertainty ledger; the capacity-aware Next Kindest Step; Living Plan adaptation through “This is too much”; transparent provider and fallback attribution; focused backend tests, Flutter analysis, widget testing and a production web build.
What we learned
Intelligence alone does not create trust. Trust comes from admitting uncertainty, asking permission, protecting boundaries and giving people the right to stop. I also learned that the best first action is not always the largest or most impressive one. Sometimes it is simply the step a person can take without their hands beginning to shake. One step. Then another. But not alone.
What's next
NotAlone can grow through permissioned plugins for trusted-contact handoffs, emergency information, safe destinations, parked-car memory and journey breadcrumbs. The AI may prepare, explain and recommend these actions. The person will continue to decide what is actually executed. NotAlone cannot promise that nobody will ever feel lost again. It carries a different promise: when that moment arrives, we can build something that stays beside them until the first step becomes possible. Doing good does good. In both directions. Links Live demo: https://web-nine-wheat-83.vercel.app Source code: https://github.com/rthgit/notalone-openai-build-week Demo video: https://www.youtube.com/watch?v=V93QOew-ADU
NotAlone — A Dynamic AI Swarm for Human Situations
Your situation. Your boundaries. A plan built around you.
NotAlone turns a complex human situation into one manageable, reversible next step. Before planning, it creates a Human Agency Contract from the person's capacity, support preferences, privacy boundaries, and consent. It then builds the smallest useful team of agents for that exact need, challenges uncertain claims, and adapts the plan when the person says, “This is too much.”
Live demo
- App: https://web-nine-wheat-83.vercel.app
- API health: https://backend-mu-rose-49.vercel.app/build-week/health
- Track: Apps for Your Life
- License: MIT
- Primary Codex
/feedbackSession ID:019f5d92-1d08-75d3-a69d-74a38baa13f3
What is included
- Human Agency Contract and privacy invariants
- Dynamic Situation Model and situation-specific agents/subagents
- Independent execution events with honest provider/fallback attribution
- Critic, uncertainty ledger, and Next Kindest Step
- Living Plan adaptation that preserves explicit boundaries
- Flutter voice-first experience with the NotAlone neon ring
- Deterministic fallback when no model API key is configured
- FastAPI and Flutter tests plus Vercel configuration
Runtime honesty
GPT-5.6 through Codex contributed to development during Build Week. That is
separate from the application runtime. The hosted runtime target is Groq with
openai/gpt-oss-120b; only trace-confirmed execution is labeled live. The app
remains usable in visibly labeled deterministic fallback.
No automatic emergency call, third-party contact, location sharing, or other external action occurs without explicit human approval.
How Codex and GPT-5.6 were used
Codex was the engineering collaborator throughout Build Week, not an embedded user-facing runtime. In the primary build thread, GPT-5.6 helped inspect the inherited project, separate the Build Week scope, design the typed human situation and agency models, implement the dynamic agent runtime, harden retries and fallback attribution, build focused tests, verify the Vercel deployment, and audit the claims shown to judges.
The human creator retained the product direction, safety boundaries, consent
semantics, privacy rules, accessibility priorities, and final decisions about
what NotAlone should and should not do. The hosted application currently uses
trace-attributed openai/gpt-oss-120b through Groq or a clearly labeled
deterministic fallback; it does not claim that GPT-5.6 is the production
runtime.
Run locally
Backend
cd backend
python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt
python -m uvicorn app.main:app --host 127.0.0.1 --port 8001
Without an API key the backend intentionally uses deterministic fallback. For
the hosted Groq target, copy .env.example to .env and configure it locally;
never commit secrets.
Frontend
cd frontend
flutter pub get
flutter run -d web-server --web-hostname 127.0.0.1 --web-port 8080 `
--dart-define=OBSERVER_API_URL=http://127.0.0.1:8001
Verify
cd backend
python -m pytest tests -q
cd ..\frontend
flutter analyze
flutter test
flutter build web `
--dart-define=OBSERVER_API_URL=https://backend-mu-rose-49.vercel.app
Judge flow
- Open the public demo or load the demo situation.
- Confirm Do not contact anyone in the Human Agency Contract.
- Select Build around me.
- Review the single Next Kindest Step.
- Select This is too much.
- Confirm that the next step becomes smaller and the privacy boundary remains.
- Expand How the AI team worked to inspect execution attribution.
Roadmap — Permissioned Life Plugins
Future plugins can prepare a trusted-contact handoff, emergency information, safety-aware map destinations, parked-car memory, journey breadcrumbs, translation, documents, reminders, and qualified-help handoffs. Every plugin is off by default, visible when active, time-bounded, exportable, and deletable. No silent tracking or autonomous external action is part of the roadmap.
Additional implementation and submission evidence is available under docs/.
Analysis
View
Metric
- 3
Figures cover GitHub contributors during the hackathon window. A co-authored commit counts in full for each author, so per-member totals add up to more than the whole-team figures.
Technology
- DartIn code
- FastAPIIn code
- HTMLIn code
- PythonIn code
- VercelClaimed
4 of 5 appear in the indexed code. 1 claimed on Devpost could not be matched to code, which may simply mean the tool leaves no trace in the repository.
AI coding agents
No AI coding agent signals were found in this repository.
Detected from committed agent config files and commit authorship. Absence of a signal is not proof an agent was unused.
Codebase size
Source size
223 KB
Source files
30
Counts recognized source files only; vendored directories, binaries and lockfiles are excluded, so this is smaller than the repository on disk.
Repository
rthgit/notalone-openai-build-week
47 files · 675 KB · @ ae154a0
Structure
Interface
12 files · 26%Screens, components and styles rendered to the user.
API & routing
2 files · 4%Request entry points: routes, handlers and controllers.
Application logic
8 files · 17%Domain rules, services and shared utilities.
Supporting
Layers are inferred from where files sit in the tree, not from reading the code. A project that names its directories unconventionally will read oddly here — open the file browser to check anything the diagram implies.
Languages
- Python50%
- Dart33%
- Markdown16%
- YAML1%
- HTML1%
Share of indexed source by file size. Binary and vendored files are excluded.
Dependencies
backend/requirements.txt
pypi · 5- fastapi
- httpx
- pydantic
- python-dotenv
- uvicorn[standard]
Declared in the repository’s manifests at the indexed commit. A declared package is not proof it is used, and runtime dependencies are listed first.
This project’s features have not been analysed yet.
Export this project's context (description, README, evidence, key source files) to chat with an AI agent elsewhere.