sema-core

Project Overview

Sema Code Core is an event-driven AI coding assistant core engine that provides reliable and pluggable intelligent processing capabilities for building code assistant tools. It supports core capabilities such as multi-agent collaboration, Skill extension, and Plan mode task planning, and can be quickly integrated into various AI programming tools.

View Documentation

Core Features

Feature	Description
Natural Language Instructions	Directly drive programming tasks through natural language
Permission Control	Fine-grained permission management ensures safe and controllable operations
Subagent Management	Multi-agent collaboration with dynamic sub-agent scheduling based on task types
Skill Extension Mechanism	Plugin architecture to flexibly extend AI programming capabilities
Plan Mode Task Planning	Decomposition and execution planning of complex tasks
MCP Protocol Support	Built-in Model Context Protocol service to support tool extension
Multi-Model Support	Compatible with Anthropic, OpenAI SDK, and LLM APIs from major vendors

Scenarios

• IDE / Editor Plugin Development — Provides low-level AI capability encapsulation for editors, allowing developers to focus on UI interaction without self-developing complex large model scheduling and tool calling logic.

• Enterprise Internal R&D Tools — Private deployment + permission control, adapting to enterprise-owned models and security specifications. Out-of-the-box toolchain avoids building AI programming infrastructure from scratch.

• Vertical Domain Intelligent Workflow — Decomposes complex engineering tasks (migration, refactoring, documentation) into automated processes. Multi-agent collaborative execution replaces manual processing of repetitive coding work.

• Academic Research & Agent Prototype Verification — Provides a lightweight Agent experimental environment for academic institutions and independent researchers, supporting flexible combinations of toolchains and agent strategies, allowing researchers to focus on algorithmic innovation.

Use Cases

VSCode Extension

Sema Code VSCode Extension is a VSCode intelligent programming plugin based on the Sema Code Core engine.

SemaClaw

SemaClaw is a general-purpose engineering harness for building personal AI agents.

Skill Web App

A Skill web application based on Sema Code Core, integrating Agent Skill Browser / Creator / Playground demo.

Quick Start

1. Create a project and install dependencies

mkdir my-app && cd my-app
npm init -y
npm install sema-core

2. Download the example file

Download quickstart.mjs to the my-app directory, then modify the following two configurations:

const core = new SemaCore({
  workingDir: '/path/to/your/project', // Target repository path for the Agent to operate on
  ...
});

const modelConfig = {
  apiKey: 'sk-your-api-key', // Replace with your API Key
  ...
};

For more model configuration options, see Model Management

3. Run

node quickstart.mjs

Cross-Language Integration Reference README.md

Development

# 1. Install dependencies
npm install

# 2. Build
npm run build

License & Third-Party Dependencies

sema-core is released under the MIT License.

All production dependencies are pulled from the npm registry under permissive licenses (MIT / ISC / BSD / Apache-2.0 / BlueOak-1.0.0 / 0BSD); no copyleft (GPL / AGPL / LGPL) code is introduced. The full component list, license summary, and external service terms are tracked in THIRD_PARTY_NOTICES.md. Resolved versions are pinned via package-lock.json for reproducible installs and license traceability.

When calling third-party model or service APIs (Anthropic, OpenAI, OpenAI-compatible endpoints, MCP servers) through this library, the embedding application is responsible for complying with the corresponding provider's terms of service.