ARCHITECTURE

How it works.

Upload data or connect any API. Both paths end at the same runtime. Your credentials encrypted, your AI agents connected, every call logged.

DATA → API

File in. API out.

Your file becomes a queryable REST API. Types detected automatically. No code generation, no deployment step.

YOU
Upload CSV, JSON, or Excel
RUNTIME
Detect format, infer types, create table
RESULT
Live API URL with filter, sort, search, export
TERMINAL
$ curl -F file=@data.csv api.liteio.dev/upload {"name":"products", "rows":2847, "api":"https://api.liteio.dev/d/products"} $ curl "api.liteio.dev/d/products?category=electronics&sort=-rating&limit=2" {"data": [{"name":"Pro Headphones", "rating":4.9}, {"name":"USB-C Hub", "rating":4.7}]}
CONNECT → PROXY

Store credentials once. Call any endpoint.

Your credentials are encrypted and stored. When you call through the proxy, we decrypt them, inject auth headers, and stream the upstream response back.

YOUR APP
/proxy/cn_a8f3/v1/models
API HUB
Decrypt credentials, inject auth headers
OPENAI
Response streamed back to your app
TERMINAL
$ curl -X POST api.liteio.dev/connections -H "Authorization: Bearer $TOKEN" \ -d '{"name":"OpenAI","base_url":"https://api.openai.com", "auth_type":"bearer","auth_config":{"token":"sk-proj-..."}}' {"id":"cn_a8f3e1b2"} $ curl api.liteio.dev/proxy/cn_a8f3e1b2/v1/models -H "Authorization: Bearer $TOKEN" 200 OK 42ms {"data": [{"id":"gpt-4o"}, {"id":"gpt-4o-mini"}]}
AI AGENTS

Claude and ChatGPT use the same pipeline.

Connect your AI via MCP. It can query your uploaded data, browse API endpoints, and make authenticated calls directly from the conversation.

CLAUDE
"What are the top-rated electronics?"
MCP
JSON-RPC 2.0, same auth, same proxy
DATA / API
Query your data or call upstream APIs
SECURITY

Your credentials are safe.

Encrypted at rest. Decrypted only at call time. Deleted from memory after the response.

Encrypted credentials

AES-GCM-256 with a unique key per account. Decrypted only during the proxy call, then garbage collected.

Tenant isolation

Every resource scoped to your account. Every query enforces it. No cross-tenant access possible.

Audit log

Every proxy call recorded with method, status, and duration. No request or response body stored.

No passwords

Authentication via Ed25519 public-key cryptography. No password database. No reset flows.

Built with TypeScript, Hono, SQLite, and the Web Crypto API. One codebase. One runtime.

Start building.

Upload your first file or connect your first API.

# How It Works Upload data or connect any API. Both paths end at the same runtime. Your credentials encrypted, your AI agents connected, every call logged. ## Upload Data
You API │ │ │ POST /upload │ │ (CSV, JSON, Excel) │ │ ─────────────────────────→ │ │ │ Detect format │ │ Infer types (1000 rows) │ │ Create typed table │ │ │ ← {name, rows, api_url} │ │ │ │ GET /d/products?price:gt=50&sort=-rating │ ─────────────────────────→ │ │ │ Parameterized SQL query │ ← {data: [...]} │
Your file becomes a live REST API. Types detected automatically. Filter, sort, search, paginate, all from query parameters. Schema and column statistics available at /schema and /stats. ## Connect Any API
You API Hub Upstream │ │ │ │ POST /connections │ │ │ {name, base_url, creds} │ │ │ ─────────────────────────→ │ │ │ │ Encrypt credentials │ │ ← {id: "cn_a8f3"} │ │ │ │ │ │ GET /proxy/cn_a8f3/v1/models │ │ ─────────────────────────→ │ │ │ │ Decrypt credentials │ │ │ Inject auth headers │ │ │ ────────────────────────→│ │ │ ← response │ │ ← {data: [...]} │ │
Store credentials once. Call any endpoint through the proxy. We inject your auth headers and stream the response back. Your client code never touches a secret. Supports bearer tokens, API keys, basic auth, and OAuth2. Upload an OpenAPI spec to auto-index all endpoints. ## AI Agents
Claude / ChatGPT API Hub Upstream │ │ │ │ tools/call: api_call │ │ │ {connection_id, method, │ │ │ path} │ │ │ ─────────────────────────→ │ │ │ │ Same proxy pipeline │ │ │ ────────────────────────→│ │ │ ← response │ │ ← {status, body, │ │ │ duration_ms} │ │
Connect Claude or ChatGPT via MCP (JSON-RPC 2.0). AI agents use the same encrypted proxy pipeline. They can query your uploaded data, browse API endpoints, and make authenticated calls directly from the conversation. ## Security Your credentials are encrypted with AES-GCM-256 and decrypted only when we make the API call. After the response is sent, the plaintext is deleted from memory. - Encrypted at rest: each account gets a unique derived encryption key - Tenant isolation: every resource scoped to your account, enforced on every query - Audit log: every proxy call recorded with method, status, and duration - No passwords: authentication via Ed25519 public-key cryptography ## Built With TypeScript, Hono, SQLite, Web Crypto API. One codebase. One runtime. ## Links - Developer Guide: Quickstart and code examples - API Reference: Full endpoint documentation - Marketplace: Browse and connect popular APIs