Documentation for the AI-Native executive coaching platform
An AI-native advisory team framework with seven specialized advisors, five operational modes (advise, coach, plan, assist, execute), human/expert-in-the-loop controls, behavioral adaptation, and multi-agent orchestration.
Route questions to specialized advisors automatically. The router uses LLM classification to pick 1-4 advisors, selects a lead, and synthesizes their perspectives. Users can also explicitly select advisors or use @mentions.
Five operational modes control how advisors respond: Advise (recommend), Coach (Socratic), Plan (action items), Assist (draft artifacts), Execute (take actions). Modes can be set explicitly or auto-detected from message intent.
Thumbs up/down feedback drives an AI-analyzed, human-approved learning loop. When negative feedback accumulates, the system proposes behavioral directives that are injected at runtime without redeploying agents.
Execute mode uses configurable tool trust levels (auto/confirm/blocked) with session-level batch approval. External experts can review conversations and post inline comments that are incorporated into agent context.
OAuth/SSO authentication via Auth.js v5. Project-scoped workspaces with per-user deployment targets, knowledge bases, and behavioral directives.
Adapter-based knowledge system supporting standalone SQLite FTS5 or Busibox RAG. Per-project scoping with a common knowledge pool and explicit sharing.
The platform wraps in Electron for a native desktop experience.
| Guide | Description |
|---|---|
| Architecture | System design, data flow, and component overview |
| Advisors | Each advisor’s expertise, modes, and the QA Judge |
| Admin Console | Deployment targets, setup wizard, MCP authentication |
| Deployment | How to deploy agents to CMA or Busibox |
| Knowledge Base | Knowledge provider interface and project scoping |
| API Reference | All REST API endpoints |
| Development | Local setup, testing, contributing |