Siada CLI

简体中文 | English

This repository contains Siada CLI, a command-line AI workflow tool that provides specialized intelligent agents for code development, debugging, and automation tasks.

Siada CLI is context-aware of your project, computing environment, and cross-session persistent memory, offering consistent, personalized, and continuously improving intelligent assistance across Terminal, Feishu (Lark), ACP clients, and Web interfaces.

Current latest version: v1.7.21

---

What's New in v1.7.21

1. Runs Everywhere, Plugs Into Anything

• True cross-platform — fully supported on macOS, Linux, and Windows, with a consistent experience on all three. Not a POSIX tool with a compatibility shim: shell execution is platform-native (pexpect PTY on Unix, PowerShell / cmd detection on Windows), and terminal theme detection, notifications, IPC (Unix socket vs. named pipe), daemon spawning, ripgrep binaries, and the Node UI bootstrap all have dedicated per-platform paths.
• Native Feishu (Lark) bot — a built-in bot transport with streaming cards and typing indicators, so you can drive Siada from chat without writing any glue code. See Remote Control Guide.
• ACP-native — ships a standalone Agent Client Protocol server (siada-acp), so Siada drops into any ACP-compatible client (Zed, Kiro, vscode-acp, …) as a first-class agent. Terminal, chat, editor, and Web all talk to the same core.

2. Squeezes the Most Out of Every Model

• On par with the best on frontier models — on Claude and GPT families, Siada CLI matches leading coding tools such as Claude Code and Codex on real-world code generation and bug fixing.
• Deeply tuned for domestic/open models — dedicated engineering for GLM-5.2, DeepSeek V4 Flash / Pro, Kimi, and Qwen: reasoning-content replay across multi-turn tool calls, per-model thinking-parameter mapping, per-model context and token budgets, and hallucination/loop suppression for fast models that invent tool names or repeat identical calls.
• One interface, any provider — unified provider layer over LiteLLM, plus custom endpoints for private deployments.

3. Long-Horizon Execution and Self-Evolving Memory

• Long-horizon task execution — automatic task-complexity judgment, spec-driven Research → Plan → Act pipelines, sub-agent orchestration with clean context windows, checkpoints, and auto-compaction keep goals on track across task chains of hundreds of steps.
• Self-evolving memory — cross-session persistent memory combined with a structured fact store (entity extraction, holographic/HRR vector retrieval, trust scoring, and contradiction detection) plus a background review-and-update pipeline. Memory is not just an append-only log: it is scored, corrected, and pruned over time.
• Proactive automation — run Siada as a background daemon with scheduled tasks so it keeps working, and keeps improving, between your sessions.

---

What Siada CLI Can Do

• Code Generation — Create new features, web interfaces, and refactor code with multi-language support
• Bug Fixing — Automatically identify, analyze, and fix code defects in large codebases
• Long-Horizon Task Execution — Tackle complex, multi-step development tasks through spec-driven execution and sub-agent orchestration
• Self-Evolution — Persistent memory, fact store, and scheduled proactive tasks keep the assistant continuously improving
• Host Management — Remote management and processing of your local machine and documents
• Multi-Entry Support — Terminal (interactive/non-interactive), Feishu bot, ACP clients, or Web UI

---

Installation & Update

System Requirements

Requirement	Details
OS	macOS, Linux, Windows
Python	3.12 – 3.13
GCC	11+ (macOS / Linux)
Package Manager	uv

Install

Step 1: Install uv

macOS / Linux:

curl -LsSf https://astral.sh/uv/install.sh | sh

Windows (PowerShell):

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

Step 2: Install Siada CLI (same command on all platforms)

uv tool install --force --python python3.12 --compile --with pip siada-cli@latest

If siada-cli is not found on PATH after installation:

uv tool update-shell

Update

uv tool upgrade siada-cli

Uninstall

uv tool uninstall siada-cli

---

Developer Mode Installation

1. Prerequisites: Ensure you have Python 3.12+ and Poetry installed.

2. Clone and Install:

   git clone https://github.com/liauto-siada/siada-cli.git
   cd siada-cli
   poetry install
   

3. Build UI (required before first run):

   cd siada_cli_ui && npm ci && npm run build:all && cd ..
   

4. Run CLI:

   # Method 1: Run with Poetry
   poetry run siada-cli
   
   # Method 2: Activate virtual environment (recommended)
   source $(poetry env info --path)/bin/activate
   siada-cli
   

---

Configuration

Quick Model Setup

On first launch, Siada CLI will guide you through an interactive setup — simply select your provider, choose a model, and enter your API key. No manual configuration needed to get started.

For advanced configuration (config file, environment variables, CLI flags), see the User Manual.

For custom external models (private deployments), see External Model Configuration.

---

Usage Modes

Interactive Mode (continuous conversation)

siada-cli

Non-Interactive Mode (script / CI integration)

siada-cli --no-interactive "Fix the null pointer exception in src/main.py"

Feishu (Lark) Bot

Connect Siada CLI to your Feishu workspace and send tasks from chat — natively supported, with streaming replies. See Remote Control Guide.

ACP Mode (editor / third-party clients)

Expose Siada over the Agent Client Protocol so any ACP-compatible client can use it:

siada-acp

Or launch the bundled UI over ACP:

siada-cli --acp

---

Key Features

Memory System & Fact Store

Cross-session persistent memory lets Siada remember your preferences, codebase context, and past decisions. On top of it, a structured fact store extracts named entities, retrieves via holographic (HRR) vectors, tracks trust scores with feedback, and detects contradictions — so knowledge is refined rather than merely accumulated.

Checkpoint

Save and restore agent states at key decision points, giving you full control over long-running tasks.

Skills

Define reusable instruction sets and workflows as SKILL.md files that extend Siada's capabilities for your specific domain or team conventions. See Skills Documentation.

MCP Support

Connect to external tool servers via the Model Context Protocol (MCP) to extend Siada's toolset.

Daemon & Scheduled Tasks

Run Siada as a background daemon to execute proactive, scheduled automation tasks.

Instruction Context (Rules)

Place siada_rule.md or .siadarules/ files in your project or home directory to set persistent AI behavior rules without repeating yourself in every conversation.

Auto Compaction

Intelligently compress conversation history to stay within model context limits without losing critical context. See Compaction Configuration.

Platform-Aware Tooling

Tool execution adapts to the host OS instead of assuming a single environment: PTY-based shell on macOS/Linux, PowerShell/cmd auto-detection on Windows, platform-specific ripgrep binaries, native notifications, and OS-appropriate IPC and daemon handling. Prompts also carry the detected OS so the model generates commands that actually run on your machine.

---

Documentation

Document	Description
User Manual	Full usage guide, commands, and configuration
Skills System	Create and manage reusable AI skills
External Model Config	Configure private/custom model endpoints
Remote Control (Feishu)	Control Siada via Feishu messages
Compaction Configuration	Context compression strategy settings
Contributing Guide	How to contribute to this project

---

Contributing

We welcome contributions to Siada CLI! Whether you want to fix bugs, add new features, improve documentation, or suggest enhancements, your contributions are greatly appreciated.

Please read our Contributing Guide before submitting:

• Project vision and development goals
• Directory structure and development guidelines
• Pull request guidelines and best practices
• Code organization principles

---

Acknowledgements

Siada CLI is built upon the foundation of numerous excellent open source projects. We extend our deepest respect and gratitude to their contributors.

Special thanks to:

• OpenAI Agents SDK — foundational agent framework
• LiteLLM — unified LLM provider interface
• OpenHands ACI — agent-computer interface tooling
• Gemini CLI — inspiration for terminal-native AI agent design

For a complete list of open source projects and licenses, see CREDITS.md.

---

License

Distributed under the Apache 2.0 License.

Disclaimers

See disclaimers.md.

---