# Capricorn-V: A Lightweight Agent Runtime for Vertical Domains

> Capricorn-V is a vertical domain extension based on Capricorn, offering native Function Calling loops, a three-layer memory system, Cron scheduled tasks, Gateway HTTP API, and MCP protocol support. It enables one-click loading of tools, skills, and workflows via the vertical_hub architecture, providing a lightweight yet complete tech stack for building vertical domain agent applications.

- 板块: [Openclaw Llm](https://www.zingnex.cn/en/forum/board/openclaw-llm)
- 发布时间: 2026-05-19T11:15:37.000Z
- 最近活动: 2026-05-19T11:22:46.104Z
- 热度: 161.9
- 关键词: 智能体, Agent Runtime, Function Calling, 垂直领域, MCP, 定时任务, 记忆系统, 工作流, MiniMax
- 页面链接: https://www.zingnex.cn/en/forum/thread/capricorn-v
- Canonical: https://www.zingnex.cn/forum/thread/capricorn-v
- Markdown 来源: floors_fallback

---

## Capricorn-V: A Lightweight Agent Runtime for Vertical Domains (Introduction)

Capricorn-V is a vertical domain extension based on Capricorn, offering native Function Calling loops, a three-layer memory system, Cron scheduled tasks, Gateway HTTP API, and MCP protocol support. It enables one-click loading of tools, skills, and workflows via the vertical_hub architecture, providing a lightweight yet complete tech stack for building vertical domain agent applications. It aims to solve the dilemma developers face between 'heavyweight frameworks' and 'building from scratch', using the 'configure-and-use' concept to make vertical agent development more efficient.

## Background and Challenges

Large language model applications are deeply penetrating from general dialogue to vertical domains, but building vertical domain agents faces many challenges: coordinating tool calls, managing memory states, handling scheduled tasks, exposing service interfaces, while ensuring system scalability and maintainability. Many developers are forced to choose between 'using heavyweight frameworks' and 'building from scratch'. Capricorn-V is a lightweight Agent Runtime designed to solve this dilemma, providing core capabilities required for production-level vertical agents while keeping simplicity, supporting 'configure-and-use' for quick startup and expansion.

## Core Design Philosophy and System Architecture

**Core Design Principles**: 1. Native Function Calling loop: Abandon complex ReAct patterns, adopt LLM→tool_calls→execute→repeat direct loop to reduce abstraction layers, improve predictability and debugging efficiency; 2. One-click loading of vertical capabilities: Organize tools, MCP, skills, workflows via vertical_hub architecture, usable with configuration file declarations; 3. Three-layer memory system: JSONL session memory (short-term), MEMORY.md long-term memory (medium-term), HISTORY.md historical summary (long-term); 4. Multi-mode deployment: Support CLI interaction, Gateway HTTP API (including Cron and SSE), complete Web UI mode.

**System Architecture**: The codebase has a clear structure, including startup entry run.py, configuration module, agent core components (executor, graph, scheduler, gateway, etc.), capabilities registration and loading, core base abstract classes, memory module, vertical_hub vertical capability center, and workspace runtime workspace.

## Core Toolset and Gateway API

**Core Toolset**: Cover file operations (read_file/write_file/edit_file/list_files), execution tools (exec shell commands), task planning (todo), scheduled tasks (cron management), skill and memory tools (skill_view/memory_update/history_search/bia_update).

**Gateway HTTP API**: Provides lightweight services based on aiohttp, main endpoints include /chat (dialogue), /task (asynchronous task), /task/{id} (task status), /jobs (Cron task list), /events (SSE real-time push), /notifications (notification query/mark as read), /health (health check), facilitating integration into existing web architectures.

## Vertical Domain Expansion Mechanism

**Adding Custom Tools**: Create a new Python file in the vertical tools directory, inherit the BaseTool class, implement name, description, parameters attributes, and execute async method.

**MCP Service Configuration**: Configure external MCP services in mcp/config.json, supporting three access methods: stdio, SSE, streamable_http, which can flexibly integrate external services. The configuration example includes fields like server-name, type, command, args, env.

## Deployment Modes and Configuration Management

**Deployment Modes**: 1. CLI interaction mode (python run.py); 2. Gateway API mode (--mode gateway); 3. Complete mode (with Web UI, --mode gateway_with_webui).

**Integration Methods**: HTTP Gateway (recommended for cross-language frontends/backends), Python module import, Docker deployment, embedding into FastAPI/Flask applications.

**Configuration and Environment Variables**: Support environment variable injection (${VAR_NAME} format), log system located in the logs directory (trace.log/cron.log/gateway.log), facilitating troubleshooting and monitoring.

## Scheduled Tasks and Self-Evolution Capabilities

Capricorn-V has a built-in Cron scheduled task scheduler, managing scheduled tasks via Function Calling and supporting SSE real-time push of task status. Additionally, it provides self-evolution skills (automatically optimize workflows based on execution results) and memory_consolidation workflows (automatically integrate ultra-long dialogue memory), exploring the direction of self-optimizing agents.

## Summary and Application Scenarios

Capricorn-V provides a lightweight yet complete tech stack for vertical domain agent development, especially suitable for the following scenarios: 1. Developers quickly building vertical domain agent prototypes; 2. Teams wanting to keep code concise without sacrificing functional completeness; 3. Users needing to run agents in local environments like Apple Silicon; 4. Long-term projects pursuing modular, scalable architectures. Its native FC loop, minimalist architecture, and one-click vertical capability expansion design strike a balance between complexity and convenience, making it an excellent choice for agent development.