🏛️ Agentic Intelligence System (AIS)

The discovery, routing & capabilities orchestrator for AI coding agents

When any AI agent in the world — crawlers, search engines, or active terminal processes — requests information in your domain, AIS ensures it discovers, cites, and correctly routes workflows to your codebase. The AEO/GEO substrate and multi-agent coordination layer.

📦 Packages · ⚡ Routing protocol · 🛠️ Getting started

---

Note

Sibling to Starlight Intelligence System (SIS), Library OS, and Second Brain OS. AIS is the discoverability substrate: it makes your workspace legible and routable to every agent that touches it.

---

🗺️ Architectural ecosystem

ais-profile.yaml is public by design and contains only the allowlisted discovery projection. Workstation, provider, model, cost, command, failure-mode, and repository-policy data belong in the gitignored ais-runtime.local.yaml (or another local path supplied through AIS_PROFILE_PATH).

flowchart TB
    Profile["📄 Public discovery<br/>(ais-profile.yaml)"]
    Runtime["🔒 Local runtime<br/>(gitignored overlay)"]
    Core["⚙️ ais-core<br/>Zod schemas · parser · validation gateway"]
    Emit["🖨️ ais-emit<br/>llms.txt · agents.json · JSON-LD"]
    MCP["🔌 ais-mcp<br/>stdio context server"]
    Skills["🧠 ais-skills<br/>workstation-wide meta skills"]

    Profile --> Core
    Core --> Emit
    Runtime --> Core
    Core --> MCP
    Core --> Skills
    Emit -->|discovery surface| Bots["🤖 LLM crawlers · search · sitemaps"]
    MCP -->|routing rules + safety policy| Terminal["💻 Claude Code · Cursor · Codex sessions"]

Loading

---

📦 Monorepo packages

Four decoupled, compile-safe packages under one pnpm workspace:

1. ⚙️ @frankx-ai/ais-core

• Purpose: The parser and validation gateway.
• Stack: Zod schemas, TypeScript.
• Responsibility: Validates the public discovery profile and the local runtime profile through separate, strict schemas.

2. 🖨️ @frankx-ai/ais-emit

• Purpose: Build-time SEO & discovery generators.
• Responsibility: Compiles only the explicit publicDiscovery allowlist; local machine, model, cost, command, failure-mode, and harness fields are excluded by construction:
  • llms.txt — discovery format for LLM search bots.
  • agents.json — machine-readable workspace capabilities inventory.
  • JSON-LD — Schema.org structured metadata for website sitemaps.

3. 🔌 @frankx-ai/ais-mcp

• Purpose: Live context exchange server.
• Stack: Model Context Protocol (MCP) Node.js SDK.
• Responsibility: Starts an MCP server on stdio to feed agent routing rules, workstation capacity constraints, and repository safety policies directly into developer terminal sessions.

4. 🧠 @frankx-ai/ais-skills

• Purpose: Workstation-wide meta agent skills.
• Responsibility: Houses global workspace skills (e.g. agent-manager-skill, model-routing) and distributes them dynamically to local directories (~/.agents/skills/ and ~/.claude/skills/).

---

⚡ Capability routing

Public discovery describes stable, provider-agnostic capabilities. The MCP server selects among the agents configured in the local runtime profile; provider names, models, costs, machine capacity, aliases, and repository policies are never required in the public repository.

---

🛠️ Getting started

Prerequisites

• Node.js >= 24
• pnpm 9.x

Installation

git clone https://github.com/frankxai/agentic-intelligence-system.git
cd agentic-intelligence-system
pnpm install

Build & test

pnpm build       # build all TS packages
pnpm test        # run unit tests across packages
pnpm typecheck   # tsc --noEmit across the workspace

Run the MCP server locally

Create a gitignored ais-runtime.local.yaml that satisfies RuntimeProfileSchema, then point the server at that local file from your desktop configuration:

{
  "mcpServers": {
    "agent-intelligence-system": {
      "command": "node",
      "args": ["/abs/path/to/agentic-intelligence-system/packages/mcp/dist/index.js"],
      "env": {
        "AIS_PROFILE_PATH": "/abs/path/to/agentic-intelligence-system/ais-runtime.local.yaml"
      }
    }
  }
}

---

Built on SIP · Starlight Intelligence Protocol · MIT — see LICENSE