ActionInbox AI

Localhost-first MVP for turning messy business inputs into reviewed tasks, risks, follow-up drafts, owner summaries, and exports.

Structure

apps/api   FastAPI + SQLite backend
apps/web   Vite React frontend
data       Local SQLite database and upload storage
packages   Prompt seeds and shared project assets
specs      Implementation notes derived from the spec kit

Backend

cd apps/api
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
uvicorn main:app --reload --port 8000

Frontend

cd apps/web
npm install
npm run dev

Verification

cd apps/api && pytest
cd apps/web && npm run build

Analyzer Runtime

The app supports two analysis runtimes:

• deterministic: local Python analyzer, no network or credentials.
• openai: OpenAI Agents SDK analyzer with schema-bound output.

Runtime selection is controlled by environment:

ACTIONINBOX_ANALYZER_MODE=auto        # auto | deterministic | openai
ACTIONINBOX_OPENAI_MODEL=gpt-5.4-mini # optional model override

auto uses the OpenAI analyzer only when runtime credentials are already present in the process environment; otherwise it falls back to deterministic analysis and records a warning. No secrets are stored by the app.

Evaluation

Use the Evaluation view or API endpoint to compare deterministic and live OpenAI analysis on the same inbox item without creating tasks, drafts, approvals, or board entries:

POST /api/inbox/{id}/evaluate-analysis

The response includes provider scores, task previews, warnings, quality deltas, and a recommended provider.