Agentic Patterns ✨

A friendly collection of reusable agentic patterns and workflows that make LLMs feel faster, smarter, and more reliable.

Goals: ⚡ Speed · 🛡️ Reliability · 💸 Cost · 👀 Observability · 🧩 Composability

---

Overview

This repo hosts practical agentic patterns you can copy, adapt, and compose into your own systems — built from scratch with minimal dependencies. Each pattern focuses on real product concerns and avoids heavy frameworks.

• ⚡ Lower latency with parallelism and early stopping
• 🛡️ Higher reliability with structured fallbacks and judging
• 💸 Lower cost by spending where it matters (previews vs full answers)
• 👀 Better observability with logs/metrics/hooks
• 🧩 Composable pieces you can mix-and-match

Philosophy (clear and simple)

• 🏗️ Barebones-first: Implementations from scratch. No LangChain or similar orchestration frameworks.
• 🧰 SDK-first: Prefer official SDKs and thin clients (e.g., OpenAI Agents SDK), or direct HTTP calls when simpler.
• 🪶 Minimal deps: Small, readable modules over complex abstractions.
• 🔬 Transparent control: Explicit routing, retries, and evaluation; no hidden magic.
• 🔌 Portable: Patterns you can lift into any stack or infra.

Non-goals (what this repo is not)

• 🚫 A framework or all-in-one runtime
• 🚫 A plugin marketplace or prompt-in-a-box solution
• 🚫 Vendor lock-in; patterns should generalize across providers

Who is this for?

• Engineers who want to understand and control agent behavior end-to-end
• Teams who prefer direct SDKs over large orchestration layers
• Builders who value clarity, reliability, and production-minded patterns

Contributing

Contributions are welcome! Please:

• Keep modules self‑contained with a clear README.md and minimal dependencies
• Prefer small, composable abstractions over frameworks
• Include a quickstart and a realistic example per module

If you’re unsure where to start, open a discussion or a small draft PR — we’re friendly 🙂

License

MIT License — see LICENSE.