SciTools

SciTools builds software for deep code comprehension, analysis, and onboarding.

For over two decades, SciTools has helped developers, security teams, and organizations understand large, complex codebases — especially those that are unfamiliar, legacy, or safety-critical.

Our flagship products focus on answering hard questions about code:

How does this system really work?
Where are the risks, dependencies, and design boundaries?
How can new developers become productive faster?

Models we ship

These are the models Understand and Onboard offer in their model menus. We test every model on this list, and we ship the exact file we tested. A result never carries over from one quantization to another.

Model	Made by	Size	Understand	Onboard	Download
Qwen3-1.7B Q4_K_M	Alibaba (China)	1.1 GB	✓		Download
Qwen3.5-2B Q4_K_M	Alibaba (China)	1.3 GB	✓	✓	Download
Granite 4.1 3B UD-Q4_K_XL	IBM (US)	2.2 GB	✓ (default)		Download
Gemma 4 E2B Q4_K_M	Google (US)	3.1 GB	✓	✓ (default)	Download
Gemma 4 E4B QAT UD-Q4_K_XL	Google (US)	4.2 GB	✓	✓	Download
Qwen3.5-9B Q4_K_M	Alibaba (China)	5.7 GB	✓	✓	Download
Gemma 4 12B Q4_K_M	Google (US)	7.1 GB	✓	✓	Download
gpt-oss-20b Q4_K_M	OpenAI (US)	11.6 GB	✓		Download
Qwen3.6-35B-A3B UD-Q4_K_XL	Alibaba (China)	22.4 GB	✓	✓	Download
gpt-oss-120b Q4_K_M (2 parts)	OpenAI (US)	63 GB		✓	Part 1 · Part 2

Understand also ships the bge-small-en-v1.5 embedding model (f16, 68 MB) for search. Download

Some of our customers need US-made models: Granite, Gemma and gpt-oss are made in the US.

Which model should I use?

For Understand, see Choose a model in Understand's help. It compares every model in Understand's menu, with our test results and the memory each needs, and we keep it current.
For Onboard, see the SciTools/OnBoard model card.

Other files in this organization

These repositories also hold older models and files we used in testing. Our products no longer offer them, but they are still here for anyone who uses them:

SciTools/Qwen3: Qwen3 0.6B, 4B, 8B and 14B
SciTools/granite: Granite 4.0 Micro, Granite 4.0 1B and Granite 4.1 3B Q4_0, the earlier default
SciTools/Llama3.2: Llama 3.2 1B, 3B and 8B
SciTools/gemma and SciTools/OnBoard: Gemma 4 E4B Q4_K_M, the earlier E4B file

Each file's SHA-256 checksum is shown on its file page.

What We Build

Understand
A static code analysis and comprehension platform that provides rich insight into code structure, dependencies, metrics, and architecture across many programming languages.

Onboard
An AI-assisted web tool designed to help developers explore and understand codebases more effectively by combining static analysis with local language models.

Our Approach to AI

We focus on local, transparent, and developer-controlled AI.

Models run locally whenever possible
We say where every model came from and publish the exact file we tested
No hidden fine-tuning or prompt manipulation
Designed to integrate with real engineering workflows

The models hosted here are provided to support SciTools products as well as other GGUF-compatible tools.

Who Uses SciTools

SciTools products are used by:

Software engineers and architects
Security and assurance teams
Organizations working with large or long-lived codebases
Teams operating in regulated or high-assurance environments

Learn More

Website: https://scitools.com
Documentation & Support: https://support.scitools.com

SciTools is a small, engineering-focused company.
We value clarity, correctness, and tools that earn trust over time.