What's shipped. What's coming. No marketing — just facts.

Drop-in OpenAI/Anthropic proxy with prompt compaction, two-tier cache, and opt-in model cascade. Returns standard responses with X-Cartie-* headers showing real savings.

Anomaly detection, idle resource flagging, RI maximizer, multi-cloud spend tracking, carbon-cost optimizer, and 21 more. Live data, real probes against your billing exports.

For every cost line in your cloud bill: who introduced it. Engineer leaderboard + Code→Cost drill-down + per-feature + per-customer rollups. Demo accounts get deterministic sample data; real customers populate via GitHub webhook + OTel hooks.

Per-repo tier (T1-T4) + 14-day shadow mode default. Customer controls how aggressively CARTIE may act. Foundation for the PR auto-fixer.

GET /api/me/guarantee-status returns real subscription_paid vs savings_booked math. Dashboard widget tracks status. Stripe live billing ships once Mercury clears.

Every action you or CARTIE takes is logged with timestamp, actor, target, IP, user-agent. Customer-readable, CSV-exportable, 365-day retention by default.

DELETE /api/me/data wipes everything you own and returns a signed deletion receipt. Tested end-to-end.

POST /api/llm-proxy/keys/:id/rotate atomically issues a new secret and invalidates the old one. Audit-logged.

Live HTTP probes against MongoDB, AI providers, email + currency vendors. 90-day uptime sparkline computed from real probe data.

RFC 9116-compliant /.well-known/security.txt. Disclosure contact: Harinathmekala@cartieai.com. 1-day acknowledgment, 5-day initial assessment.

SSE responses are captured chunk-by-chunk on first call and replayed on identical-prompt hits. Unlocks the 80% of production traffic that previous v2 cache bypassed. Same X-Cartie-Cache-Hit header, full streaming UX preserved.

Public page walking enterprise reviewers through Shadow Mode → T3 → T2 → T1 tiered autonomy with concrete what-happens-when-CARTIE-acts examples per tier.

GET /api/savings/ledger pulls realized AI savings + estimated cloud recommendation savings into a single timeline with per-engine breakdown. Powers the Dashboard banner.

Code Tracer is built — now we attach a real GitHub App that opens PRs at the configured tier (respecting shadow mode + excluded paths). Awaiting App registration under the founder's account.

Engine #27 already books AI savings. Cloud savings get the same treatment — post-merge billing comparison attributing realized $ to specific PRs.

When a cost anomaly fires in Slack, CARTIE automatically runs the Code Tracer against the customer repo and either opens a PR (T1/T2/T3) or proposes one (T4).

Vanta / Drata readiness work begins once we sign the first enterprise pilot. Audit completion target: 4-6 months from kickoff.

Optional thin wrapper that auto-injects base_url + X-Cartie-Key. MIT licensed, source on GitHub for full auditability.

For air-gapped enterprises. Helm chart + docker-compose, MongoDB or Postgres backend, BYOK cloud bill exports.

After SOC 2 Type 2 (typically Series A timeline).