Best AI tools for ai agents

Automations you can host yourself and inspect.

Workflow automation you can self-host. Connect apps and add AI steps you can see.

A visual agent or knowledge bot you can ship as an API.

Workspace for agent workflows and knowledge bots you can ship as an app or an API.

Multi-agent jobs with roles and a runtime.

Framework for multi-agent jobs, plus a hosted product for teams that need control.

Agent logic in code, when a visual builder is not enough.

Open source libraries, including LangGraph, for building agents as code.

Drag-and-drop agent flows on your own machine.

Open source visual builder for chat flows and agents you can self-host.

A coding agent that already works on a real repository.

Anthropic coding agent that reads a repo, edits files, and runs commands.

Shared team agents on company docs and tools.

Hosted agents for a team, connected to company tools such as Slack and docs.

Connecting the apps you already use, without hosting anything.

Connects apps into automations, including steps that call an AI model.

Handing off repeat work without writing an agent framework.

Hosted agent you connect to email, Slack, and other tools to do repeat work.

Everyday agent-style tasks when you already live in ChatGPT.

OpenAI chat app for questions, drafts, and agent-style tasks.

A large open-weight model for coding and agents, if you can run or host it.

Meta open-weight multimodal model for coding and agent work. Weights are public under a license with commercial limits.

A fast model API when your agent needs a cheap, quick call.

Diffusion language model API from Inception Labs, with a 1 million token context window.

A morning brief and background tasks from Anthropic's research preview.

Anthropic research-preview agent that sends a morning brief, takes tasks by message, and keeps working in the background.

Agents you host yourself, on a model and machine you choose.

Open-source managed agent platform you can self-host, for running agents on any model and machine.

Letting a model operate Android apps from the screen.

Android app that reads the screen through the accessibility service and taps and types in other apps.

A multimodal model API with a 1 million token context.

MiniMax multimodal model for text, image, and video input, with a 1 million token context.

A 1 million token model on the Kimi API.

Moonshot's first K3.1 model, described as 3T-class, with a 1 million token context. Weights are not downloadable yet.

An MCP server for a website that does not already have an API.

Turns a website that has no API into a hosted API or MCP server for agents, from a prompt.

Gemini models and Managed Agents, including Deep Research, from the terminal.

Google's experimental CLI for Gemini models, media generation, and Managed Agents. The CLI is free. API use is billed.

Computer use and visual reasoning from ChatGPT Work or the API.

OpenAI model for computer use, visual reasoning, and writing. Available in ChatGPT Work, Codex, and the API as gpt-6.1-terra.

High-volume agent tasks, once early access exists.

Mirendil's first model, pretrained from scratch, with a context window of up to 10 million tokens. Access is early-access only.

A desktop agent that runs an open-weight model on your computer.

Free, MIT-licensed desktop app that works on your local files with open-weight models, and it can run fully offline.

Letting an agent sign in without seeing the password.

Lets an approved agent sign in with your 1Password vault without seeing the password or one-time code.

A frontier Gemini model via the API or Google Antigravity.

Google's frontier Gemini model, in the Gemini API, the Gemini app, and Google Antigravity.