Code with AI. Prove nothing leaked.
Confide is an IDE that routes every AI completion through a hardware-isolated Trusted Execution Environment (TEE) on NEAR AI Cloud. Your code, prompts, and context never leave your boundary — and every reply ships with a cryptographic attestation receipt you can verify in under 30 seconds.
Built on the open NEAR AI stack.
| Surface | Status | Description |
|---|---|---|
| Landing page | ✅ shipped | Dark marketing site — hero w/ NEAR pill, features, metrics, pricing tiers w/ real margins, silver MagicRings CTA, footer |
| Workspace shell | ✅ shipped | Left sidebar (Chat / Playground / Browse Models / Usage / Settings) + top bar with profile menu + "NEAR TEE connected" badge |
/chat workspace |
✅ shipped | Chat panel + live Scanner panel; per-message attestation receipts (model, TEE, hash, latency, tokens); 5-model picker with real NEAR IDs |
/playground |
✅ shipped (mock CVM) | Paste GitHub URL → spawn confidential sandbox. Status pane stepper; trust-model explainer; real Phala/Azure spawn is P1 |
/browse-models |
✅ shipped | NEAR-only catalog (40 models) with real provider logos for Anthropic / Google / DeepSeek / OpenAI / Qwen / GLM; search + category + creator filters |
/usage |
✅ shipped | Square stat grid (Agent Runs / Sessions / Tokens / USD) backed by per-request UsageEvent log; time-range tabs; per-day bar chart; per-model leaderboard |
/settings/profile |
✅ shipped | Avatar, linked emails with Primary/OAuth/Verified pills, danger zone |
| GitHub OAuth | ✅ shipped | Connect GitHub from /playground, browse your repos, one-click import to spawn a sandbox. End users never see secrets — see md/09-github-oauth.md |
| Backend service | ✅ shipped | Fastify on :4000 — /v1/chat (NEAR proxy), /v1/usage/*, /v1/models, /v1/attestation/report, /v1/sandbox/*, /v1/auth/github/*, /v1/github/repos |
| Token usage tracking | ✅ shipped | Backend appends UsageEvent per chat to data/usage.jsonl; cost computed from NEAR per-million pricing; /usage/summary returns totals + by-day + leaderboard in one call |
| Real CVM spawn (Phala adapter) | ⏳ next | Replace the mock spawn with real TDX VMs — see md/08-playground-design.md phased plan |
| In-VM editor + terminal | ⏳ next | Monaco editor + xterm.js terminal in browser, WS bridge to confide-agent Go binary inside the CVM |
| Frontend → backend chat handoff | ⏳ next | One-file change: point frontend/app/api/chat/route.ts at backend:4000/v1/chat so usage events persist to disk |
| Privacy upgrade (E2EE / BYOK direct) | later | Cut Confide's proxy out of the plaintext path — see md/plan.md §10 |
| Pricing / Stripe | later | Marketing copy only; pricing math is in the landing tiers |
| VS Code extension | later | Real CDE clone after the web IDE is solid |
YOUR LAPTOP CONFIDE (this repo) NEAR AI CLOUD
┌─────────────────┐ ┌───────────────────┐ ┌──────────────────────┐
│ │ │ │ │ ┌──────────────────┐ │
│ Code editor │ ── prompt ──▶ │ Next.js IDE │ ── HTTPS ──▶ │ │ Intel TDX TEE │ │
│ + AI plugin │ │ /api/chat │ (sk-... key) │ │ + H100 CC GPU │ │
│ │ ◀── reply ── │ │ ◀── receipt ── │ │ + GLM-4.6 / … │ │
└─────────────────┘ └───────────────────┘ │ └──────────────────┘ │
│ │ │ │ │
▼ ▼ │ ▼ │
┌──────────┐ ┌──────────────┐ │ Attestation │
│ Provider │ │ Scanner UI │ │ (signed by TEE) │
│ sees │ │ shows the │ └──────────────────────┘
│ your raw │ │ receipt │
│ code 😱 │ │ per message │
└──────────┘ └──────────────┘
┌──────────────────────────┐ ┌──────────────────────────┐ ┌──────────────────────────────┐
│ USER DEVICE │ │ CONFIDE SERVER │ │ NEAR AI CLOUD │
│ │ │ │ │ │
│ ┌────────────────────┐ │ │ ┌────────────────────┐ │ │ ┌────────────────────────┐ │
│ │ Confide IDE │──┼───┼─▶│ /api/chat │──┼───┼─▶│ Gateway │ │
│ │ Next.js + React 19 │ │ │ │ Next.js route │ │ │ │ OpenAI-compatible │ │
│ └─────────▲──────────┘ │ │ └─────────┬──────────┘ │ │ └───────────┬────────────┘ │
│ │ │ │ │ │ │ ▼ │
│ ┌─────────┴──────────┐ │ │ ┌─────────▼──────────┐ │ │ ┌────────────────────────┐ │
│ │ Scanner UI │◀─┼───┼──│ Attestation parser │ │ │ │ Intel TDX TEE │ │
│ │ attestation panel │ │ │ └────────────────────┘ │ │ │ + H100 CC GPU │ │
│ └────────────────────┘ │ │ │ │ │ ├─ GLM-4.6 FP8 │ │
└──────────────────────────┘ └──────────────────────────┘ │ │ ├─ DeepSeek V3.1 │ │
│ │ ├─ GPT-OSS 120B │ │
│ │ └─ Qwen3 30B │ │
│ └───────────┬────────────┘ │
│ ▼ │
│ ┌────────────────────────┐ │
│ │ Attestation service │ │
│ │ signs receipt │ │
│ └────────────────────────┘ │
└──────────────────────────────┘
Developer Confide IDE /api/chat NEAR TEE Scanner
│ │ │ │ │
│── type ───────▶│ │ │ │
│ prompt │ │ │ │
│ │── POST /api ───▶│ │ │
│ │ /chat │ │ │
│ │ │── POST /v1/ ──▶│ │
│ │ │ completions │ │
│ │ │ (Bearer sk-…) │ │
│ │ │ │ decrypt in │
│ │ │ │ TEE, │
│ │ │ │ run model, │
│ │ │ │ sign receipt │
│ │ │◀── reply + ────│ │
│ │ │ receipt │ │
│ │◀── reply +──────│ │ │
│ │ receipt │ │ │
│◀── render ─────│ │ │ │
│ reply │ │ │ │
│ │── render ───────┼────────────────┼─────────────▶│
│ │ receipt card │ │ │
▼ ▼ ▼ ▼ ▼
| Layer | Tool |
|---|---|
| Framework | Next.js 16 (App Router) |
| UI | React 19, Tailwind v4, framer-motion |
| Inference | NEAR AI Cloud — cloud-api.near.ai |
| Models | GLM-4.6 FP8, DeepSeek V3.1, GPT-OSS 120B, Qwen3 30B A3B |
| Enclave | Intel TDX + Nvidia H100 Confidential Compute |
| Hosting | Vercel (planned) |
cd frontend
cp .env.example .env.local # paste your NEAR_API_KEY
npm install
npm run dev # http://localhost:3000The landing page renders without a key. The /chat workspace will require
NEAR_API_KEY once it's built.
near boom/
├── README.md ← this file
├── plan.md ← full MVP plan
├── near-ai-cloud-api.md ← NEAR API reference
└── frontend/
├── package.json
├── .env.example
└── app/
├── layout.tsx
├── page.tsx ← landing
├── globals.css
└── components/
├── Navbar.tsx
└── ui/
├── reveal.tsx
├── animated-wave.tsx
├── metrics-section.tsx
└── powered-by.tsx
| Without TEE | With Confide on NEAR |
|---|---|
| Provider sees your raw code | Provider sees only ciphertext |
| You trust the SaaS log policy | You hold a signed receipt |
| Defense, finance, healthcare can't use AI coding tools | They can — verifiably |
| One vendor lock-in per model | One API, four frontier models |
| "Trust us" | "Verify us" |
- ✅ Landing page (dark theme, animated, full sections, mobile-responsive, NEAR branding)
- ✅ Confide logo + browser favicon
- ✅
/chatworkspace UI (two-pane: chat + Scanner) - ✅
/api/chatserver route proxying tocloud-api.near.ai/v1/chat/completions - ✅ Scanner panel renders real fields when NEAR returns them; graceful stub when key/credits absent
- ✅ Model picker wired to NEAR's real model IDs (Claude Opus 4.7, GLM 5.1, DeepSeek V3.1, GPT-OSS 120B, Qwen3 30B)
- ✅ Profitable pricing tiers based on NEAR's actual cost-per-million-token math
- ✅ Full NEAR AI Cloud docs captured in
md/(quickstart, models, reasoning, private inference, gateway/TLS attestation, complete API endpoint reference) - ✅ Hosted brand logos on
/modelsfor Anthropic, Google, DeepSeek, OpenAI/GPT, Qwen/Alibaba, GLM/Z.ai (Moonshot + Black Forest Labs still on letter-badge fallback) - ✅ Silver/white MagicRings WebGL backdrop on the landing final CTA
- ⏳ Live TEE attestation against a real response (blocked on NEAR account credits — key set but
HTTP 402 no_limit_configured) - ⏳ "Verify receipt" button that calls
/v1/attestation/reportper message - ⏳ Local audit log (browser localStorage for receipt history)
- ⏳ Live model list (call
GET /v1/model/listinstead of hardcoded picker)
See md/plan.md for the full roadmap and md/ for the NEAR AI Cloud docs we've captured (quickstart, models, reasoning, private inference, gateway + TLS attestation, API endpoint reference).
Private — © 2026 Ayush Singh. All rights reserved.