AI-Native Coach Platform

Documentation for the AI-Native executive coaching platform

---

Project maintained by jazzmind Hosted on GitHub Pages — Theme by mattgraham

Admin Console

Overview

The admin console (/admin) provides a UI for managing deployment targets, deploying advisors, and connecting MCP servers. Configuration is per-user – each authenticated user manages their own deployment targets and API keys.

Setup Wizard

Access at /admin/setup. A 4-step flow:

Step 1: Choose Target Type

• Claude Managed Agents – deploy to Anthropic’s managed cloud
• Busibox – deploy to a self-hosted Busibox instance

Step 2: Enter Credentials

• CMA: Anthropic API key
• Busibox: Host URL + API key

Step 3: Validate

Credentials are tested against the target API:

• CMA: sends a minimal Haiku request
• Busibox: hits the /api/health endpoint

Step 4: Deploy

All 8 agents (7 advisors + QA Judge) are deployed to the target. The wizard shows progress and reports success/failure.

Target Detail Page

Access at /admin/targets/[id]. Shows:

Status Summary

• Deployment status (deployed, error, deploying, unconfigured)
• Number of deployed agents
• Last deployment timestamp

MCP Server Connections

Manage OAuth connections for MCP servers:

• Notion – opens Notion’s OAuth flow at mcp.notion.com/mcp
• Slack – opens Slack’s OAuth flow at mcp.slack.com/mcp
• Google Workspace – requires Claude connector (deferred)

Connection status is tracked per target so you can have different MCP configurations for different deployment targets.

Deployed Agents

List of all agents with:

• Agent name and ID
• Version number
• Health status (click “Check Status” to poll the target API)

Actions

• Redeploy All – re-deploys all agents (updates existing, creates missing)
• Check Status – polls the target for agent health
• Delete – removes the target configuration (does not archive deployed agents)

Config Store

The admin console stores configuration in SQLite with encrypted credentials:

Tables

config – key-value settings store (per-user)

deploy_targets – deployment target configurations (per-user)

• Credentials encrypted at rest using AES-256-GCM
• Key derived from CONFIG_ENCRYPTION_KEY env var or a machine-specific fallback
• Agent state (IDs, versions) stored alongside for update/redeploy

mcp_connections – MCP server auth state per target

Deployment Adapters

The adapter pattern (lib/deploy/adapter.ts) defines a common interface:

interface DeployAdapter {
  validate(config): Promise<{ valid, error? }>
  deploy(coaches, config): Promise<DeployResult>
  status(config, agentState): Promise<TargetStatus>
  teardown(config, agentState): Promise<{ success, error? }>
}

CMA Adapter

Ports the logic from deploy.py to TypeScript:

1. Creates/reuses an environment
2. Deploys base agents (QA Judge) first
3. Deploys dependent agents with callable_agents references
4. Prepends date/time preamble to system prompts

Busibox Adapter

Pushes INSTRUCTIONS.md content to the Busibox Agent API:

1. Validates connection via /api/health
2. POSTs each advisor to /api/agents
3. Checks status via GET /api/agents