@m6d/cortex-server

Multi-agent AI chat server built on Hono + Bun. Supports MSSQL persistence, MinIO file storage, Neo4j knowledge graphs, and AI streaming with tool execution.

Usage

import { createCortex } from "@m6d/cortex-server";

const cortex = createCortex({
    database: {
        type: "mssql",
        connectionString: process.env.DB_CONNECTION_STRING!,
    },
    storage: {
        endPoint: "localhost",
        port: 9000,
        useSSL: false,
        accessKey: "minioadmin",
        secretKey: "minioadmin",
    },
    auth: {
        jwksUri: "https://your-auth-provider/.well-known/jwks.json",
        issuer: "https://your-auth-provider/",
    },
    model: {
        baseURL: "https://api.openai.com/v1",
        apiKey: process.env.MODEL_KEY!,
        modelName: "gpt-4o",
    },
    embedding: {
        baseURL: "https://api.openai.com/v1",
        apiKey: process.env.EMBEDDING_KEY!,
        modelName: "text-embedding-3-small",
        dimension: 1536,
    },
    neo4j: {
        url: "bolt://localhost:7687",
        user: "neo4j",
        password: "password",
    },
    agents: {
        assistant: {
            systemPrompt: "You are a helpful assistant.",
            tools: {},
        },
    },
});

const server = await cortex.serve();

export default {
    fetch: server.fetch,
    websocket: server.websocket,
};

cortex.serve() is async — it runs database migrations on startup.

Each key in agents becomes a route prefix (e.g. assistant → /agents/assistant/...). Agents can define per-agent systemPrompt, tools, backendFetch, loadSessionData, resolveRequestContext, and lifecycle hooks (onToolCall, onStreamFinish).

Requirements

• Bun >= 1.0.0