What Is LLM Explorer and What Can You Find on It?

LLM Explorer (llm-explorer.com) is a directory site for open-source language models and AI agents, operated by Extractum. It indexes 63,069 open-source LLMs and 150 AI agents in one place, so you can search, filter, and compare models by category, benchmark, or hardware fit instead of tracking down each project's repo separately. It's most useful when you already know roughly what you need — a model that fits your VRAM, an agent framework, a quantized build — and want a shortlist rather than a single recommendation.

What the site covers

The directory splits into two main sections:

Open-source language models — 63,069 indexed models, with 14,683 quantized variants and 23,379 models that fit within 8 GB of VRAM. Models are grouped into browsable categories rather than one flat list.

AI agents — 150 agents, each listed with type, GitHub stars, forks, and license. The directory separates "most popular" from "recently updated," which matters because star count and maintenance activity often point to different projects.

Model categories you can browse

The category list is the fastest way in if you have a constraint in mind:

There's also a dedicated list of TheBloke's models, which is useful if you specifically want that publisher's quantizations.

What each agent entry tells you

Agent listings use a consistent set of fields, so you can compare across projects:

• Agent name and author — e.g. AutoGPT (Significant Gravitas), Dify (LangGenius), browser-use
• Type — Ready-to-Use, Platform, Framework, or SDK
• Stars and forks — a rough signal of adoption and community activity
• License — MIT, Apache-2.0, GPL-3.0-or-later, AGPL-3.0, Elastic-2.0, or "Other"

The type field is the one to read first. A "Ready-to-Use" agent is something you run; a "Framework" or "SDK" is something you build with. Mixing those up wastes time.

How to use it for a real decision

Say you have 8 GB of VRAM and want a code model. A practical path:

1. Open the "LLMs Fit in 8GB RAM" category to cut the list to models your hardware can actually run.
2. Narrow further with "Code Generating Models."
3. Cross-check candidates against the LMSYS ChatBot Arena ELO list if you want a quality signal beyond the category tags.
4. If you need a specific runtime format, filter by GGUF, GPTQ, or EXL2 — the site tracks 14,683 quantized models, so most popular bases have a quantized build listed.

For agents, start from "Recently Updated" rather than "Most Popular" if you plan to depend on the project long-term. A high star count on a stalled repo is a common trap; the recently-updated list surfaces projects like Hermes Agent, OpenCode, and Gemini CLI that are still moving.

Limits worth knowing

The site is an index, not a benchmark lab. Star counts, forks, and licenses are pulled from project metadata, and category tags like "fit in 8GB" are directory classifications — verify against the model card before committing, since quantization level and context length both change actual memory use. The directory also doesn't publish pricing information, so treat it as a discovery and comparison layer, not a source for cost or commercial terms. For licensing questions, follow the license link on each entry to the original repo.