Snap
emnlmn.github.io- Category
- AI Tools
- Rank
- No. 3774Tools index
Previous survey · No. 3608 ·
- Pricing
- Open Source
- Type
- TOOL
- Use case
- Models: Train & Run
- Interfaces
- CLI · Web
- GitHub
- 10 stars
- Latest release
- v0.6.2
- Date
About
Snap is a local decision engine that classifies, ranks and decides about an input in sub-second time on a personal machine. It does not generate text. It reads logits in a single forward pass, and supports four classes of questions. On a Mac, brew install and snap serve pull the snap1-2b model from Hugging Face and start a local server with a web playground. The 2B model scores 0.643 zero-shot accuracy on typed decisions, and the weights are open under the MIT license.
At a glance
Snap turns text into typed decisions by reading the model scores for a fixed set of answers. It is worth evaluating if your pipeline needs labels or flags and currently spends time generating and parsing a text response.
Good fit
Developers routing tickets or classifying events into known options who can measure mistakes on labeled examples.
Consider something else when
Skip it if you need explanations or have to discover new answer categories as you go.
Start here
Read the sample support-ticket request, then assemble a small set of synthetic tickets with known labels. Include an ambiguous ticket and see whether your pipeline has a sensible fallback when the ranking is close.
Cost and requirements
The project lists MIT licensing and downloads for macOS, Linux and Windows. Local use needs a compatible model and enough compute to run it.
Limits to know
A distribution over supplied answers can be confidently wrong or leave out the right category. Calibrate any threshold on representative examples before using it to trigger actions.
Related work
- Newt (alternative)
Sources checked 2026-11-05. Source-based assessment.
Tags
Tech Stack
Media

Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.