feat: initial implementation of Oliver AI Cost Tracker
Complete Phase 1 implementation:
Backend (FastAPI + MongoDB + Celery):
- Core: config, DB with indexes, JWT security, API key auth middleware
- Models: org hierarchy (workspace/team/project), user mirror, pricing,
usage events/rollups, budgets, alert log, audit log
- Services: pricing engine (LiteLLM/YAML/override priority), budget check
with preflight, email alerts at 50/80/100%, analytics aggregations,
audit logger
- API routes: public (preflight/record/upsert), admin CRUD, pricing
management, budget management, analytics (summary/timeseries/breakdown/pivot),
Microsoft SSO auth
- Celery tasks: daily LiteLLM price sync with change notifications,
daily rollup aggregation, 5-minute alert evaluator
- Pricing catalogue: ElevenLabs + Google Cloud TTS in models.yaml
SDK (oliver-cost-tracker Python package):
- CostTracker client with httpx + exponential backoff (3 retries)
- SQLite outbox with 30s background flusher (never blocks AI pipeline)
- Estimators: token/char estimation per provider
- BudgetExceeded / CostTrackerUnavailable exceptions
Frontend (React 18 + Vite + TypeScript):
- Dashboard with KPI cards, daily cost timeseries, top-model/top-user charts
- Pivot Explorer with multi-dim row/col selection + stacked bar chart + table
- Admin pages: Workspaces, Pricing (with LiteLLM sync + override), Budgets
(with live spend bar), API Keys (show-once), Users (mirror), Audit Log
- Microsoft SSO login flow
Infra: docker-compose.yml (mongo + redis + api + celery worker + beat + frontend)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2f070ce503