- Category
- Developer Tools
- Rank
- No. 3813Tools index
Previous survey · No. 3622 ·
- Pricing
- Open Source
- Type
- TOOL
- Use case
- Data Processing
- GitHub
- 14 stars
- Latest release
- v1.6.0
- Date
About
Solvi is a Python runtime where a model proposes answers and the developer's own code decides. It uses typed questions, hard rules, quotes checked against the source text, escalation with a guaranteed error budget, and replayable traces. Because checks make the final decision, outputs can be audited and reproduced, and it works with LLMs, custom classifiers, or small local models.
At a glance
Solvi is a Python library for decisions that combine model judgments with explicit checks. It is worth trying when a system needs to abstain or send a case to a person if a required fact cannot be verified.
Good fit
Developers who can express decision rules in code and want a trace of the facts and checks behind each answer.
Consider something else when
Use plain Python rules when all inputs and decisions are already deterministic.
Start here
Build a synthetic document decision with one required field. Omit that field and verify that a confident model guess cannot override the failed check.
Cost and requirements
Requires Python 3.11 or newer. Optional model and serving backends add dependencies and can carry separate compute or provider costs.
Limits to know
The quality of the decision depends on the checks and evidence you define. A trace can explain how a result was reached even when the policy itself needs improvement.
Sources checked 2026-11-05. Source-based assessment.
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.