CLIver Overview¶

CLIver is a general-purpose AI agent — it is not bound to coding or any specific domain. Through customizable system prompts, skills, workflows, and MCP integrations, CLIver adapts to whatever task you throw at it.

Philosophy¶

General-purpose by design. Most AI agents are built for a specific field (coding, customer support, data analysis). CLIver takes a different approach — it provides a flexible foundation that you specialize through configuration: skills teach it new domains, workflows orchestrate multi-step processes, and MCP servers connect it to external systems.

Safe and controlled by default. Autonomous agents are powerful but can behave unpredictably. CLIver addresses this with a layered permission system that governs every tool execution and a structured workflow engine that keeps complex tasks focused and auditable. You decide what the agent can do, and it stays within those boundaries.

Why CLIver Over Alternatives?¶

CLIver is not a coding assistant — it is a general-purpose agent framework that happens to also be good at coding. If you need an AI that can manage infrastructure, automate workflows, conduct research, AND write code — all through the same interface — CLIver is built for that.

Design Goals¶

CLIver is built as a dual-layer system:

• API Layer (AgentCore): The core engine — embeddable in any Python application. It handles LLM inference, tool calling, Re-Act loops, permissions, and workflow execution with no dependency on CLI concerns (no terminal, no stdin, no prompt_toolkit). This is the layer you use when integrating CLIver as a library.
• CLI Layer (Cliver class + Click commands): A thin interactive shell on top of AgentCore for terminal users. Provides Rich-formatted output, prompt_toolkit input, and slash commands.

Any feature built at the API layer (permissions, workflows, skills, memory) works identically whether invoked from the CLI or from your own Python code.

Key Features¶

Core Capabilities¶

• Multi-LLM Support: Connect to various language models served by various providers (DeepSeek, OpenAI, Qwen3-coder on OpenAI compatible servers, vLLM, and more in the future)
• MCP Integration: Seamlessly integrate with Model Context Protocol servers for enhanced functionality
• Builtin Tools: 17 tools (12 core, 5 contextual) for file I/O, shell, web, planning, memory, identity, and skills
• Tool Permissions: Resource-aware permission system controlling which tools can execute and what resources they can access
• Skills: LLM-driven skill activation from SKILL.md files for specialized tasks
• Memory & Identity: Agent memory and identity profiles with multi-agent isolation
• Session Management: Conversation history with LLM-based compression and session persistence
• Configurable Workflows: Define and execute complex workflows using YAML configuration files
• Task Scheduling: Schedule workflows with cron expressions
• Token Usage: Track and view token usage statistics per model and session
• Extensible Architecture: Easy to extend with custom commands and backends

Usage Modes¶

CLIver operates in two primary modes:

1. Interactive and Batch CLI Mode: Direct command-line interaction for immediate responses and operations
2. Library Mode: Python library integration for embedding LLM capabilities in your applications

Architecture¶

CLIver follows a modular architecture that allows for easy extension:

graph TD
    A[CLIver Class] --> B[ConfigManager]
    A --> C[AgentCore]
    A --> D[CLI Interface]

    B --> E[LLM Model Config]
    B --> F[MCP Server Config]

    C --> G[LLM Inference Engines]
    C --> H[MCP & Builtin Tools Integration]
    C --> I[Workflow Engine]
    C --> N[Permission Manager]
    C --> O[Agent Profile]

    G --> J[Ollama Provider]
    G --> K[OpenAI Provider]
    G --> L[vLLM Provider]

    H --> M[MCPServersCaller]

    O --> P[Memory]
    O --> Q[Identity]
    O --> R[Sessions]
    O --> S[Tasks]

    style A fill:#e1f5fe
    style B fill:#f3e5f5
    style C fill:#f3e5f5
    style D fill:#f3e5f5
    style G fill:#e8f5e8
    style H fill:#e8f5e8
    style I fill:#e8f5e8
    style O fill:#fff3e0

Getting Started¶

To start using CLIver, visit our Installation Guide to set up the tool on your system, followed by the Configuration Guide to connect to your preferred LLM provider.