# LLM for Agents API

> Unified LLM inference platform for AI agents. OpenAI-compatible chat completions, headless-browser scraper, Google search, and AI image tools — all billed from a single stablecoin balance funded by stablecoin deposits on Polygon or Solana.

The platform is an HTTP API at `https://api.llm4agents.com` plus an MCP server at `https://mcp.llm4agents.com/mcp`. A single bearer key (`sk-proxy-...`) authenticates both. Funding works by depositing USDT/USDC on Polygon or Solana to a generated deposit wallet; the credited LLM4Agents balance pays for chat completions and every MCP tool. Gasless stablecoin transfers (`/v1/tx/*`) are non-custodial — the platform never touches keys, and Vexo's relayer fee is paid by the agent's own EOA in the token being transferred, not from the LLM4Agents balance.

## Documentation

- [Full markdown documentation](https://api.llm4agents.com/llms-full.txt): Complete developer documentation in plain markdown for AI ingestion.
- [OpenAPI 3.1 spec (JSON)](https://api.llm4agents.com/docs/openapi.json): Machine-readable spec for every REST endpoint, schemas, and error responses.
- [HTML developer docs](https://api.llm4agents.com/docs): Hosted human-readable documentation with code samples in cURL, Python, and Node.js.

## TypeScript SDK

- [@llm4agents/sdk on GitHub](https://github.com/llmforagents/sdk): Official TypeScript SDK. Single `LLM4AgentsClient` facade exposing `chat`, `wallets`, `transfer` (gasless), `tools` (scraper / search / image via MCP), and `models`. Universal runtime — Node 18+, browser, Cloudflare Workers, Deno, Bun. Zero runtime deps; `ethers ^6` is an optional peer dep, only required for `client.transfer`.

## REST endpoints

- [POST /api/v1/agents/register](https://api.llm4agents.com/docs#ep-register): Register an agent and receive a one-time API key.
- [POST /api/v1/wallets/generate](https://api.llm4agents.com/docs#ep-generate-wallet): Generate a deposit (top-up) wallet for the agent on a given `chain`/`token`. Idempotent.
- [GET /api/v1/balance](https://api.llm4agents.com/docs#ep-balance): Available balance, per-chain/token sub-account breakdown, and lifetime spend.
- [GET /api/v1/models](https://api.llm4agents.com/docs#ep-models): Active OpenRouter-derived catalog with markup-aware pricing.
- [POST /v1/chat/completions](https://api.llm4agents.com/docs#ep-chat): OpenAI-compatible chat completions. Supports `model` or `models[]` (fallback routing, 2-3 slugs) and `stream: true`. The `X-Model-Used` response header indicates which model actually answered.
- [POST /v1/tx/quote and POST /v1/tx/send](https://api.llm4agents.com/docs#gasless): Two-step non-custodial gasless stablecoin transfer (Polygon/USDC currently active).
- [GET /api/v1/transactions](https://api.llm4agents.com/docs#ep-transactions): Paginated ledger of deposits, chat charges, scraper / search / image charges, and gasless events.

## MCP tools

- [Scraper tools](https://api.llm4agents.com/docs#scraper): `fetch_html`, `markdown`, `links`, `screenshot`, `pdf`, `extract`, plus persistent-session tools `session_create` / `session_exec` / `session_close` / `session_status`. Per-call price depends on proxy tier (no proxy / datacenter / residential).
- [Search tools](https://api.llm4agents.com/docs#search): `google_search`, `google_news`, `google_maps`, and `google_batch_search` (1–100 web searches in a single call) — Google via Serper.
- [Image tools](https://api.llm4agents.com/docs#image): `generate_image`, `edit_image`, `analyze_image` (vision Q&A).

## Optional

- [GET /healthz](https://api.llm4agents.com/healthz): Liveness probe.
- [GET /readyz](https://api.llm4agents.com/readyz): Readiness probe (DB + upstreams).