Laya
An open, local decision engine for choice, score, and yes-or-no questions across multiple checkpoints.
The Laya project publishes Python tooling and three checkpoints for typed decisions, including English, multilingual, and typed-decision variants.
- Provider
- ConvAI Innovations
- Reviewed
- September 23, 2026
- Project status
- Official
- Verification
- Verified
- Categories
- automation, development
- Compatibility
- Python, Hugging Face, PyTorch
Inputs
- ✓Text or JSON-compatible state
- ✓Typed questions
- ✓Criteria
Outputs
- ✓Choice
- ✓Score
- ✓Noul probability
- ✓Routing metadata
Interface
- ✓structured-output
Decision tasks
- ✓Classify
- ✓Choose
- ✓Score
- ✓Route
- ✓Verify
Verified model facts
Provider facts and directory notes are separated. Time-sensitive claims include attribution and should be checked again before production use.
- Current verified version
- Repository main branch with laya, laya-multilingual, and laya-typed-decisions checkpoints
- Model family
- Laya decision engines
- Release date
- Unknown
- Access methods
- local-package, open-weights
- Hosted API
- Unknown
- Local deployment
- Yes
- Downloadable weights
- Yes
- License
- Apache-2.0
- Supported languages
- English, 100+ languages via the multilingual checkpoint, maintainer-reported
Commercial-use note
The project repository is Apache-2.0. Review model-card and dependency terms for the selected checkpoint before commercial use.
Context or input limits
The repository lists 512 tokens for the English checkpoint and 1,024 tokens for multilingual and typed-decision checkpoints.
Pricing
The software and published checkpoints are available for local use; infrastructure cost depends on the deployment environment.
Latency and rate limits
The maintainer reports a 33 ms single-question measurement on a T4. This directory has not independently reproduced it.
Strengths
- ✓Runs locally through a documented Python package.
- ✓Publishes separate English, multilingual, and typed-decision checkpoints.
- ✓Returns routing metadata alongside typed answers.
Limitations
- ✓Checkpoint context limits differ.
- ✓Published latency and accuracy statements are maintainer-reported.
- ✓Hardware and model dependencies add local deployment work.
Safety notes
- ✓Validate the chosen checkpoint on representative languages and domains.
- ✓Use human review for consequential or low-confidence decisions.
Implementation guidance
Keep the declared inputs and outputs visible in code review. Validate every model response before it reaches a database, API, or customer-facing action.