Build with AI
Agent frameworks, coding agents, MCP, IDEs, and AI app-building tools.
Build with AI.
Agent frameworks, autonomous coding tools, AI-native IDEs, and the protocols and infrastructure that connect models to real software.
Agents frequently run with elevated permissions and can execute shell commands or modify files without confirmation. Avoid granting full root or drive access — a container, VPS, or VM is strongly recommended for anything autonomous.
Agent frameworks give a language model the ability to plan, call tools, and act autonomously instead of just answering a single prompt. The space splits into single-agent frameworks built for one autonomous actor, and multi-agent orchestration frameworks designed to coordinate several agents working together on the same task.
AutoGPT
The original autonomous agent framework that sparked the agent movement, with modular architecture and forge-based agent building.
LangChain / LangGraph
The foundational library for agents, chains, and memory — LangGraph adds stateful, controllable agent orchestration on top.
- MetaGPT — multi-agent framework where agents collaborate as programmers, designers, and testers, Apache 2.0
- AWS Labs Agent Squad — flexible multi-agent orchestration with intent classification, pre-built for Bedrock and Lex
- A2A Protocol — Google's Agent2Agent open protocol for interoperability between agentic applications, donated to the Linux Foundation
- Langflow — open-source visual canvas for building and deploying agents with 1000+ integrations across frameworks
- PocketFlow — a 100-line minimalist LLM framework for lightweight agent workflows
As agent ecosystems mature, standardized protocols are emerging so agents built on different frameworks can still talk to each other, share memory, and expose tools safely.
Long-running agents need persistent memory beyond a single context window. Tools like Letta, dedicated memory-engine APIs, and local-first memory lifecycles (e.g. Ring Memory) let agents store, retrieve, and reuse context across sessions without re-processing everything from scratch each run.
These agents operate directly inside your terminal or repository — reading code, writing patches, running tests, and iterating without constant hand-holding. Most fall into either an IDE-integrated assistant or a fully terminal-native harness that manages its own subagents and memory.
OpenHands
Full-featured open-source AI software engineer, formerly OpenDevin, built for end-to-end autonomous development tasks.
Gemini CLI
Open-source agent bringing Gemini's capabilities directly into your terminal, with code generation and shell execution.
A newer layer of tooling packages entire development processes — planning, TDD, code review, spec refinement — as reusable "skills" agents can load. These sit on top of Claude Code, Codex, and other agents rather than replacing them, and are worth exploring once you've settled on a base agent.
Not every AI workflow needs custom code. Visual builders let you wire models, tools, and data sources together on a canvas, which is often the fastest path from idea to working prototype.
Beyond chat-based coding help, these are full environments and utilities built specifically around AI-assisted development — from complete IDE replacements to focused CLI utilities that plug into whichever agent you're already using.
Cursor
A full IDE built around AI-assisted multi-file editing, one of the most widely adopted AI-first development environments.
- NeMo Guardrails (NVIDIA) — programmable safety rules and dialog flows for LLM apps, integrates with LangChain/LangGraph
- Guardrails AI — input/output validation for PII, toxicity, and prompt injection with 50+ pre-built validators
- Microsoft PyRIT — automated red-teaming framework for generative AI with multi-modal attack support
- Presidio (Microsoft) — detect, redact, and anonymize PII across text and images
For teams or individuals who want a complete AI stack running on their own infrastructure — chat interface, model serving, and agent orchestration all self-hosted rather than stitched together from cloud APIs.
Several open-source projects now offer full "ChatGPT replacement" stacks with RAG, access controls, and audit logging built in — suited for teams that need an internal AI knowledge base without sending data to a third-party cloud. Worth cross-referencing with the Local AI page for model-serving setup.
Distinct from coding agents, this class of tool lives in your everyday messaging apps — Telegram, WhatsApp, Discord, Slack — and handles real-world tasks like email, scheduling, and browsing on your behalf rather than writing code.
These two are often run together rather than as competitors: OpenClaw contributes ecosystem breadth and multi-channel orchestration, while Hermes Agent contributes the persistent memory and self-improving skill layer that OpenClaw lacks out of the box. A common setup runs Hermes as a high-level planner on top of OpenClaw's tools.