Introducing All the Agents

There was no existing comprehensive list of coding agent harnesses. There were plenty of smaller collections and “awesome” lists, but nothing that provided a broad, current view of the space.

New tools are appearing constantly. Older projects become dormant or disappear. To complicate things further, several different kinds of software are described as agents even when they work in substantially different ways.

We built the resource we wanted to be able to use, and shared it as a resource for everyone.

At the time I’m writing this, All the Agents publishes 828 entries across three categories:

• 548 coding agents
• 233 multiplexers
• 47 agent SDKs

A coding agent or harness is a system that uses tools in a loop to create or modify software. Most importantly, it owns that loop. It decides what to do next, invokes a tool, evaluates the result, and continues working.

We use a reasonably simple test when classifying one: if you remove the host application, can the product still run a coding task from beginning to end? If it can, it is an agent.

A multiplexer is different. It provides a human-facing way to launch, schedule, coordinate, or otherwise manage multiple agents, but it does not do the coding itself.

An agent SDK provides the building blocks developers use to create agents. It does not ship a coding agent of its own. AutoGen, CrewAI, and LangGraph are examples of this category.

We also track more than 500 related projects that do not fit any of those definitions, including model gateways, prompt libraries, workflow packs, MCP servers, memory layers, datasets, and evaluation tools. Those projects are recorded in the public repository, but they are not published in the main catalog.

Finding the right agent

The catalog is designed to be searched and browsed, rather than read from top to bottom.

You can search by name, maker, description, license, or implementation language. You can filter the complete list by category, platform, maintenance status, and license, then sort the results by name, GitHub stars, or initial release date.

There are also dedicated views for browsing agents by:

• Platform
• Maker
• License
• Release date
• GitHub stars

Each entry combines structured information with a short explanation of why the project exists, how it works, and who uses it. Depending on what is available, that information includes supported platforms, autonomy level, maintenance status, model providers, pricing, extensibility features, source availability, and installation details.

This makes the catalog useful for more than finding a name I vaguely remember. I can use it to compare projects, find actively maintained options for a particular platform, or understand where a tool fits within the broader coding-agent ecosystem.

Keeping the catalog useful

The initial catalog was assembled from 15 aggregation lists, five rounds of GitHub topic and keyword searches, GitHub API research covering more than 1,500 repositories, and manual research into commercial products.

Every published entry is checked against primary sources. We also maintain a public categorization ledger containing the decision and a short rationale for every project we evaluated. (Some of these decisions are less obvious than they first appear.)

All the Agents is intended to remain current. New entries and corrections are welcome, and each entry page links back to its source file on GitHub. If something is missing or incorrect, you can use the repository’s issue forms to add an entry or fix an existing one.

All the Agents is live now, and the complete project is available in the Prime Radiant GitHub repository.