What is Agentier?

Agentier is a TypeScript framework for building AI agent loops. An agent loop is a pattern where an LLM model repeatedly reasons about a task, invokes tools, and uses the results to continue reasoning — until it has a final answer.

Why Agentier?

Most LLM SDKs give you a single chat completion call. Building an agent requires you to manually:

• Loop between the model and tool calls
• Parse and validate tool arguments
• Handle errors, retries, rate limits
• Manage conversation history
• Stream tokens to the user

Agentier handles all of this with a single agent.run() call.

Key Features

• Provider-agnostic — One interface for OpenAI, Anthropic, Google, Ollama, and any OpenAI-compatible API
• Type-safe tools — Define tools with Zod schemas for automatic validation and TypeScript inference
• Middleware — Plug in logging, retry, rate limiting, caching, or your own custom middleware
• Memory — Persist conversations across runs with in-memory or file-based storage
• Structured output — Get typed objects from the model using Zod schemas
• Streaming — Real-time token callbacks
• Production-ready — Timeouts, abort signals, token budgets, iteration limits

How It Works

User prompt
    ↓
┌─→ Model call (LLM)
│     ↓
│   Tool calls? ──No──→ Return response
│     ↓ Yes
│   Execute tools
│     ↓
│   Add results to conversation
└────┘

The agent loop continues until the model responds without requesting any tool calls, or a limit is reached (max iterations, token budget, timeout).