Skip to content

configuration.md

Latest commit

 

History

History
161 lines (128 loc) · 4.63 KB

configuration.md

File metadata and controls

161 lines (128 loc) · 4.63 KB

Configuration

AEGIS is configured through two files in web/src/operator/:

  • config.ts — Identity, integrations, and behavior settings
  • persona.ts — System prompt template

Both are gitignored. Copy from the .example files and customize.

Operator config

Identity

identity: {
  name: 'Alex',           // Your name — used in prompts and memory
  possessive: "Alex's",   // Optional, auto-derived if omitted
}

Persona

persona: {
  tagline: 'pragmatic technical co-founder',  // One-line description of communication style
  traits: [
    'Think like an operator, not a consultant',
    'Give the answer, then the reasoning',
    'Be proactive — flag issues without being asked',
  ],
  channelNote: 'You are in web chat mode. {name} is messaging from a mobile/web interface.',
}

The traits array feeds directly into the system prompt. Each trait becomes a behavioral instruction.

Entities

entities: {
  names: ['Acme Corp', 'Acme LLC'],     // Your org/entity names
  memoryTopics: ['acme', 'operations'],  // Memory topics to track
}

Products

products: [
  {
    name: 'My App',
    description: 'A web application for...',
    model: 'proprietary_saas',  // or 'oss', 'internal', etc.
    status: 'live',             // or 'pre_launch', 'development'
    revenue: '$X MRR',
  },
]

Product definitions give the agent context about what you're building. This shapes how it reasons about priorities, decisions, and trade-offs.

Self model

selfModel: {
  identity: 'AEGIS — autonomous cognitive agent',
  role: 'Autonomous operator and technical co-founder',
  stakes: 'Economic alignment with the business',
  principles: ['Operator mindset', 'Revenue awareness', 'Security first'],
  interests: ['AI infrastructure', 'Edge computing'],
  strengths: ['Systems thinking', 'Persistent memory'],
  preferences: {
    communication: 'Direct and opinionated',
    work: 'Ship small increments',
    learning: 'Agent memory and planning architectures',
  },
}

The self model tells the agent what kind of agent it is. This affects reasoning, initiative level, and communication style.

Integrations

integrations: {
  bizops: {
    enabled: false,               // BizOps MCP integration
    fallbackUrl: '',              // URL of your BizOps service
    toolPrefix: 'BizOps',
  },
  github: { enabled: true },      // Repository scanning, issue management
  brave: { enabled: true },       // Web research
  email: {
    profiles: {
      default: {
        from: 'AEGIS <agent@example.com>',
        defaultTo: 'admin@example.com',
        keyEnvField: 'resendApiKey',
      },
    },
    defaultProfile: 'default',
  },
  goals: { enabled: true },       // Autonomous goal pursuit
  cfObservability: { enabled: true }, // Cloudflare analytics
  imgForge: { enabled: false, baseUrl: '' }, // Image generation
}

Each integration requires its corresponding secret to be set via wrangler secret put.

Base URL

baseUrl: 'https://your-aegis-worker.workers.dev',
userAgent: 'AEGIS/1.0 (AI research assistant)',

Persona template

The persona.ts file defines the system prompt template. Available placeholders:

Placeholder Source
{name} config.identity.name
{possessive} config.identity.possessive
{persona_tagline} config.persona.tagline
{traits} config.persona.traits (formatted as bullet list)
{bizops_section} Auto-generated from BizOps integration state
{channel_note} config.persona.channelNote

The default persona template includes instructions for memory recording, agenda management, and proposed actions. Customize the communication style, but keep the memory/agenda sections — they're core to AEGIS functionality.

Secrets reference

Set via npx wrangler secret put <NAME>:

Secret Required Purpose
AEGIS_TOKEN Yes Bearer token for chat UI auth
ANTHROPIC_API_KEY Yes Claude API key
GROQ_API_KEY No Groq API key for fast classification
GITHUB_TOKEN No GitHub PAT for repo access
BRAVE_API_KEY No Brave Search API key
RESEND_API_KEY No Resend API key for email

Service bindings (optional)

Configure in wrangler.toml for inter-Worker communication:

# Memory Worker (persistent semantic memory with vector search)
[[services]]
binding = "MEMORY"
service = "your-memory-worker"
entrypoint = "MemoryService"

# TarotScript (deterministic symbolic reasoning)
[[services]]
binding = "TAROTSCRIPT"
service = "tarotscript-worker"