Honcho’s differentiator is the modeling layer — inferring an entity’s state, intent, and preferences, not just storing facts. This section covers the research behind that and the pipeline that benchmarks, trains, and serves the models.
Research
The reading list and theory behind Honcho’s approach to memory and Theory of Mind. Start with the four areas of ML for honcho, then browse the rest.
The pipeline
benchmark → trace → train → serve: Excadrill → Minccino → Machamp → Metagross. See the System Map for how it feeds back into Honcho.
See Deriver Training Pipeline for repository ownership, artifact handoffs, and the release process.
Evals
- DRIFT — our internal benchmark for modeling quality (in design).
- Public benchmarks (LongMemEval, LoCoMo, BEAM) run through Excadrill.
New to the team? See the Research onboarding track.