Agentic Intelligence System

🏛️ 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.

[](LICENSE) [](https://www.typescriptlang.org/) [](https://pnpm.io/) [](https://modelcontextprotocol.io/) [](https://github.com/frankxai/Starlight-Intelligence-System)

[📦 Packages](#packages) · [⚡ Routing protocol](#routing-protocol) · [🛠️ Getting started](#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

A single unified profile (ais-profile.yaml) drives three decoupled emitters and the live MCP server.

flowchart TB
    Profile["📄 Unified Profile Schema(ais-profile.yaml)"]
    Core["⚙️ ais-coreZod schemas · parser · validation gateway"]
    Emit["🖨️ ais-emitllms.txt · agents.json · JSON-LD"]
    MCP["🔌 ais-mcpstdio context server"]
    Skills["🧠 ais-skillsworkstation-wide meta skills"]

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

---

📦 Monorepo packages

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

1. ⚙️ [@frankx-ai/ais-core](packages/core/README.md)

• Purpose: The parser and validation gateway.
• Stack: Zod schemas, TypeScript.
• Responsibility: Parses the unified [ais-profile.yaml](ais-profile.yaml), ensuring agent specs, skill parameters, repository boundaries, and hardware capacity constraints comply with types.

2. 🖨️ [@frankx-ai/ais-emit](packages/emit/README.md)

• Purpose: Build-time SEO & discovery generators.
• Responsibility: Compiles structural documentation:
• [llms.txt](llms.txt) — discovery format for LLM search bots.
• [agents.json](agents.json) — machine-readable workspace capabilities inventory.
• [JSON-LD](jsonld.json) — Schema.org structured metadata for website sitemaps.

3. 🔌 [@frankx-ai/ais-mcp](packages/mcp/README.md)

• 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](packages/skills/README.md)

• 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/).

---

⚡ The active workstation fleet & routing protocol

AIS establishes a first-principles task-mapping system based on requirement complexity:

flowchart LR
    T["Trivial (1-3)OpenCode / Codexspeed & minimal cost"]
    M["Medium (4-6)Cursor / Clineinteractive refinement"]
    H["High (7-8)Claude Code / Antigravityautonomous TDD loops"]
    S["Substrate (9-10)DeepAgent / SIS Swarmsub-agent & delegation"]
    T --> M --> H --> S

| Complexity Tier | Target Agent | Primary LLM | Recommended Task Types | | :--- | :--- | :--- | :--- | | 1-3 | OpenCode / Codex | groq/llama-4-scout / gpt-4o | Single-file script edits, config modernizations, formatting, doc updates. | | 4-6 | Cursor / Cline | Pluggable | Interactive layouts, CSS styling, component refactoring, UI adjustments. | | 7-8 | Claude Code / Antigravity | claude-3-5-sonnet / gemini-1.5-pro | Multi-file refactors, test-driven iterations, large-context digestion. | | 9-10 | DeepAgent / SIS Swarm | Custom / Pluggable | Long-horizon multi-step planning, remote sandbox runs, agent swarms. |

---

🛠️ 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

Add the server to your Claude Code / desktop config (mcp.json), pointing at your local checkout:

{
  "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-profile.yaml"
      }
    }
  }
}

---

Built on SIP · Starlight Intelligence Protocol · MIT — see [LICENSE](LICENSE)

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: frankxai
• Source: frankxai/agentic-intelligence-system
• License: MIT
• Homepage: https://github.com/frankxai/Starlight-Intelligence-System

Install and usage instructions live in the source repository linked above.