AIX overview
Partial
Bounded task execution exists. Several document, retrieval and agent capabilities remain partial; Latera is target-only.
AIX currently runs a bounded research, synthesis and assessment workflow around authorized product context and legal evidence. Broader adaptive-agent behavior and Latera remain target work.
| Capability | Status | Implemented scope and remaining work |
|---|---|---|
| Agent execution | Partial | Durable tasks, clarification and bounded refinement exist; general adaptive routing does not |
| Context & memory | Partial | Scoped context and project-memory records exist; the full proposed memory lifecycle does not |
| Documents & drafting | Partial | Ingestion, proposals and exports exist; template discovery and native tracked-change workflows do not |
| Private retrieval | Partial | Authorized lexical context selection exists; the described reference dense/rerank pipeline is not ported |
| Legal research | Partial | Specialized judgment/legislation discovery and pinned reads exist; specialist APIs remain backlog |
| Latera | Target | No executable end-to-end Latera pipeline in this repository |
| Models & tools | Partial | Fixed typed reasoning stages and accounting exist; a general tool registry remains target |
flowchart LR Input["Task and authorized context"] --> Research["Research plan and legal sources"] Research --> Synthesis["Draft answer or proposal"] Synthesis --> Assess["Evidence checks and assessment"] Assess --> Saved["Persisted answer or reviewable proposal"]
services/product-worker/internal/capabilities/service.go coordinates the implemented sequence. TestAIXAnswerQuality, TestAIXClarificationContinuation and TestAIXAgentDocumentProposal exercise its boundaries. A passing fixture establishes behavior for that scenario, not general legal accuracy or completeness.