VennDocumentation overview#
Venn makes typed decisions over reusable embeddings. Start with a small CPU example, then choose whether a question should use described options, historical cases, or a predictor fitted on labeled examples.
Get a first result#
Follow getting started to install from the checkout and run
venn quickstart. It needs no reader or head download and uses authored vectors
to show the interface. For platform-specific Python and PyTorch setup, use
installation.
Read concepts for readers, compatible profiles, stores and families. Venn or Jev? explains when reusable state helps and where a text-reading decision model is a better fit.
Choose a workflow#
| Goal | Start here | Continue with |
|---|---|---|
| Connect an existing vector store | Store adapters | Backend usage examples, outcome decisions |
| Decide from past cases and their outcomes | Outcome stores | Small runnable example, then arena application |
| Ask Choice, Score or Noul questions about text | Serve text questions | Routing recipe, API reference |
| Fit a classifier or numeric predictor | Families | Task recipes, wine predictor, audio classifier |
| Train a local image classifier from personal photos | Hot dog image demo | Linear-family recipe, audio classifier |
| Route support tickets from resolved history | Precedent Desk | Index lifecycle and collection sweep |
| See the system before running it | Demos and videos | Complete example catalog |
Each example lists its own prerequisites. Authored-vector examples demonstrate contracts; real-reader workflows download model weights. Check the profile compatibility rules before connecting an existing store or changing readers.
Find the right kind of documentation#
- Start here: setup, the main concepts, and a comparison of approaches.
- Task guides: recipes, store adapters and implementation steps for a specific operation.
- Examples: complete applications, scripts and videos with prerequisites.
- Reference: request and response contracts, design rationale and model details.
- Results and limits: measurements, evidence, reproduction and failure boundaries.
- Project: contribution, policy and licensing information.
For exact fields and errors, go to the API reference. For interpreting reported results, use the evidence ledger and limitations; reproduction separates contract checks, geometric measurements and trained-head evaluation.
Contribute to Venn#
See contributing for development changes, bug reports and documentation improvements.