AI Detect
github.com
- Free plan
- Open source
- Windows
- macOS
- Linux
- Self-hosted
What AI Detect is
AI Detect is a free, MIT-licensed open-source tool that scores writing sentence by sentence for patterns associated with likely AI-generated text. It runs inference on your hardware, and after model download and caching it can operate offline. The default DeBERTa-v3-large model download is about 1.7 GB; a 126 MB RoBERTa ONNX model provides a lighter triage option. Diagnostics can flag missing contractions, formal verbs, and formal transitions, then offer plain-English rewrite suggestions. AI Detect is available as a command-line tool, Python package, and MCP server for Claude Desktop, Claude Code, or another MCP client. File uploads, batch scanning, and basic report exports are supported. CPU is the default, with CUDA used automatically when available. Sentences shorter than five words are skipped. The maker labels the software beta and recommends treating batch scores as signals, not definitive judgments for individual sentences.
Who it suits
It suits people checking commissioned or unpublished copy locally, including users processing batches. It also fits developers who want command-line, Python, or MCP access.
Good to know, and what to watch
- Runs text inference locally
- Works offline after model caching
- Offers lighter 126 MB model option
- Provides rewrite suggestions and diagnostics
- Supports batch scanning and MCP clients
- Default model download is about 1.7 GB
- Sentences under five words are skipped
- Beta scores are not definitive individual judgments
- No support contract or polished dashboard
Plans
Every plan and priceFree
Free plan
The verdict
AI Detect provides local, offline-capable writing analysis in several developer-friendly interfaces. Treat its beta scores as batch-level signals and plan for the model’s storage requirements.
What the maker publishes
- Accuracy caveat
- The guide reports that short factual human sentences can be falsely flagged while long conversational sentences often pass. source
- Audience
- The maker positions it for people checking commissioned or unpublished copy locally, including users handling batches of writing. source
- Benchmark claim
- Houtini says the DeBERTa detector ranked first on the RAID detection benchmark. source
- Business classification
- Companies House classifies HOUTINI LIMITED under SIC 62090, other information technology service activities. source
- Caveat
- The maker describes the software as beta and recommends treating its scores as a signal across a batch rather than a verdict on an individual sentence. source
- Command-line interface
- The CLI accepts text or files, compares two versions, emits JSON, and lets users choose model and device. source
- Company founding
- LinkedIn lists Houtini as founded in 2025. source
- Company headquarters
- LinkedIn lists Houtini’s headquarters as London. source
- Company size
- LinkedIn lists Houtini as having 2–10 employees. source
- Default model download
- The default model downloads about 1.7 GB from Hugging Face on first use and is then cached locally. source
- Detection model
- The default detector uses desklib/ai-text-detector-v1.01, based on a 304-million-parameter DeBERTa-v3-large model. source
- Diagnostics
- It flags writing patterns such as missing contractions, formal verbs, and formal transitions, and provides plain-English rewrite suggestions. source
- Hardware
- A GPU is optional: the default is CPU, with CUDA used automatically when available. source
- Hardware support
- CPU execution is the default, while CUDA is used automatically when an available GPU is detected. source
- Independent deployment status
- Glama states that this MCP server cannot be deployed through its platform. source
- Independent documentation assessment
- Glama’s page reproduces instructions for CLI commands, MCP configuration, model downloads and the five exposed tools. source
- Independent listing
- Glama lists AI Detect as a Python, local MCP server for AI text detection. source
- Input limit
- Sentences shorter than five words are skipped rather than scored. source
- Installation requirement
- The accompanying setup guide lists Python 3.9 or newer, pip and roughly 3 GB of disk space for the first download. source
- Integrations
- The MCP server can be used with Claude Desktop, Claude Code, or another MCP client. source
- Interfaces
- It is distributed as a command-line tool, Python package, and MCP server. source
- Legal entity
- UK Companies House lists the maker as HOUTINI LIMITED, company number 16683235, an active private limited company incorporated on 31 August 2025. source
- License
- The project is MIT-licensed. source
- License
- The project is released under the MIT license. source
Checked against the maker's pages on 15 Oct 2026.