Skip to main content

Open Source Strategy

Ody is built with a clear open-core model. The knowledge quality tools are free; the training pipeline and AI assistant are proprietary.

License Map

ComponentPackageLicensePurpose
Platform Core@useody/platform-coreApache 2.0Types, SQLite, providers, vector index
Detectors@useody/detectorsApache 2.05 detectors + consultant analysis + consensus
Eval@useody/evalApache 2.0Benchmarks, corpus runner, flywheel test
Export@useody/exportApache 2.0HTML report, JSONL, TRL adapter, tickets
Refine CLIody-refineApache 2.0The free wedge -- scan, detect, resolve, export
Forge CLIproprietaryProprietaryTraining pipeline, model registry
FeedbackproprietaryProprietarySignal collector, reputation, reward derivation
TrainingproprietaryProprietaryDataset registry, model registry, orchestrator
ColleagueproprietaryProprietaryWeb app, API, Slack bot, worker

Repository Structure

RepoLicenseWhat's inside
platform/Apache 2.0 (public)Core libs, detectors, eval, export, Refine CLI
forge/ProprietaryTraining, feedback, Forge CLI
colleague/ProprietaryWeb app, API, Slack bot, background worker

Why Open Source Refine?

Refine is the wedge. It provides immediate value (find broken docs in 2 minutes) with zero commitment. The health report is the viral moment -- teams share it, star the repo, and discover the rest of the platform.

The open source components are genuinely useful on their own:

  • Run ody-refine in CI to catch docs drift
  • Use detectors in your own tools
  • Export findings as tickets or training data
  • Generate health badges for your README

What Stays Proprietary

The training pipeline and AI assistant are proprietary because they are the revenue products:

  • Forge charges for managed training (GPU compute, model hosting)
  • Colleague charges per seat and consumption (the daily-use product)

Shared Contract

The @useody/* npm packages define the interface between open and proprietary:

  • @useody/platform-core -- types, document model, provider interfaces
  • @useody/eval -- benchmark definitions, evaluation framework

These packages are published under Apache 2.0 so anyone can build on top of Ody's detection and evaluation primitives.

Competitive Position

Nobody does what Ody does. The market validates adjacent problems:

CompetitorWhat they doWhat they don't do
AirweaveContext retrieval for agentsNo integrity layer
HydraDBOntology-first context graphInfrastructure, not product
MintlifyDocs formatting CINot content trustworthiness
Glean/GuruSearch enterprise knowledgeNo training loop

Ody's lane: Context integrity -- the quality gate between messy company knowledge and AI that acts on it.