VI1 modules
The 26 VI1 modules.
VI1 combines its own models with these modules. Five are wired into every VI1 API request and change answers; three power the evaluation harness; the voice and avatar modules power VI1 Voice and the 3D avatars. The rest are internal tooling.
5
Ready: in the product today
6
Beta: evaluation, voice and avatar
15
In development: callable, plan or prototype
| Module | What it does | Where it is used | Status | Commercial rights |
|---|---|---|---|---|
| calculator | Exact arithmetic tool behind vi1_calculate and checked calculations. | Wired into the API | Ready | Own |
| prompt_compiler | Builds the structured plan prompt for checked calculations. | Wired into the API | Ready | Own |
| response_observer | Checks JSON answers against the requested schema; one repair retry. | Wired into the API | Ready | Own |
| retrieval | BM25 knowledge index; search tool and auto-injected evidence. | Wired into the API | Ready | Rights review |
| closed_diagnostics | Privacy-safe provider error records (never logs prompts or keys). | Wired into the API | Ready | Own |
| offline_evaluator | Grades answers offline: exact, type-sensitive, receipt-bound. | Evaluation harness | Beta | Own |
| retrieval_evaluation | Retrieval recall, nDCG and paired bootstrap confidence intervals. | Evaluation harness | Beta | Rights review |
| synthetic_data | 2,400 records from 24 template families, split by family. | Evaluation harness | Beta | Rights review |
| training_evaluation | Train and held-out harness with strict manifest gates. | Callable | In development | Rights review |
| fixed_source_schema_guard | Schema guard for fixed data sources. | Callable utility | In development | Own |
| acceptance_ledger | Records acceptance decisions. | Callable | In development | Own |
| assignment_certificates | Statistical certificates for task assignment. | Callable | In development | Rights review |
| targeted_evidence | Evidence-selection bookkeeping. | Callable | In development | Rights review |
| protocol_selector | Picks test protocols. | Callable | In development | Rights review |
| finite_goal_planning | Bounded planner over a finite catalog. | Callable | In development | Own |
| durable_handoff | Handoff state machine. | Callable | In development | Own |
| team_controller | Multi-agent roster and lease controller. | Callable | In development | Own |
| embedding_provenance | Tracks where embeddings came from. | Callable | In development | Own |
| correction_selector | Research simulator for choosing corrections. | Research only | In development | Rights review |
| adapter_versions | Adapter version registry (prototype engine). | Prototype | In development | Own |
| distributed_training_probe | Multi-GPU training probe; needs 2+ GPUs for the real path. | Blocked on GPUs | In development | Own |
| lora_training | LoRA training plumbing; no model training done yet. | Plan only | In development | Rights review |
| quantization | Quantization plan. | Plan only | In development | Rights review |
| avatar_clock | 3D avatar and audio-driven clock demo (Three.js, CC0 assets). | Demo page | Beta | Own (CC0/MIT assets) |
| voice_backend | Voice pipeline: speech in, VI1, speech out. | Needs native app runtime | Beta | Own; provider terms |
| voice_client | Browser voice client. | Needs voice endpoints | Beta | Own |
"Rights review" marks ten bundles that need authorship or permission cleared before they are resold on their own. They are fine to use inside VI1 products.