Open Source AI Agent Licenses — What You Need to Know

Published: 2026-03-13 | Author: Nerq Research

We classified the licenses of 2,704 AI agents in the Nerq index. If you're building with open-source agents, this matters more than you think.

Why License Classification Matters

When an AI agent is integrated into your product, its license terms apply to your deployment. A coding agent with a GPL license used in your proprietary SaaS could create legal exposure. A data processing agent with no license file is legally ambiguous — you're technically using it without permission.

What We Found

We classified each license into one of five categories:

PERMISSIVE (70% of agents)

Licenses: MIT, Apache 2.0, BSD 2/3-Clause, ISC, Unlicense

These agents can be used commercially without restriction. The MIT license dominates the agent ecosystem, just as it does in the broader open-source world.

For enterprise users: This is the safe zone. PERMISSIVE agents can be used in proprietary products, deployed commercially, and modified without obligation to share changes.

COPYLEFT (14% of agents)

Licenses: GPL 2.0/3.0, LGPL 2.1/3.0, AGPL 3.0, MPL 2.0

Copyleft licenses require derivative works to be distributed under the same license. The implications depend on how tightly the agent is integrated:

• AGPL 3.0 is the most restrictive — if your users interact with an AGPL agent over a network, you may need to open-source your entire application
• LGPL is more permissive — linking to an LGPL library doesn't infect your codebase
• GPL requires distributing source for any binary that includes GPL code

For enterprise users: Proceed with caution. AGPL agents should generally be avoided in proprietary products. LGPL is usually safe. GPL depends on your distribution model.

UNKNOWN (11% of agents)

No license file found in repository

This is the most underappreciated risk. Without a license:

• The author retains full copyright
• You have no legal right to use, modify, or distribute the code
• Default copyright law applies (all rights reserved)

Many developers assume no-license means "free to use." It doesn't.

For enterprise users: Avoid agents with no license. The legal risk isn't worth it when alternatives exist.

PROPRIETARY (5% of agents)

Custom licenses, "source available" but not open source

Some agents publish their source code but under restrictive terms — no commercial use, no redistribution, or usage-based licensing.

For enterprise users: Read the actual license text. "Source available" is not "open source."

How Nerq Uses License Data

License classification feeds into the Compliance dimension (20% weight) of Trust Score v2:

License Category	Score Impact
PERMISSIVE	+10 points
COPYLEFT	+5 points
VIRAL (AGPL)	+0 points
UNKNOWN	-5 points
PROPRIETARY	-3 points

This means agents with clear, permissive licenses score higher — reflecting the reduced compliance risk for adopters.

Check Any Agent's License

curl "https://nerq.ai/v1/preflight?target=AGENT_NAME"

The response includes the license classification and compliance score.

from nerq import NerqClient
client = NerqClient()
r = client.preflight("some-agent")
print(r.license)  # "MIT"
print(r.components)  # {"compliance": 75, ...}

Top Agents by License Type

Most popular PERMISSIVE agents (MIT/Apache)

These are the agents with the highest trust scores that also have clean licensing:

Visit nerq.ai/popular for the full ranked list.

Popular agents with license concerns

We've identified several high-star agents with AGPL or UNKNOWN licenses. These are frequently used but carry compliance risk. Check them with the preflight API.

---

License data sourced from repository license files, classified using SPDX identifiers. Coverage: 2,704 agents with explicit license classification out of 204K+ indexed. Data updated daily.