简体中文 | English
Siada CLI is an AI assistant for coding, debugging, and automation. It works with your project, tools, and persistent memory to carry tasks from investigation through implementation and verification. It can also control your browser and perceive your intent from your current reading and recent actions.
News · Capabilities · Quick start · Browser preview · Documentation
Explore the browser collaboration showcase and user guide. The showcase links back to this GitHub repository.
AI's value is not in replacing people, but in making room for creativity.
Browser collaboration: the PyPI package includes the browser installer. Node.js, the proxy, and the Chrome extension are distributed as matching platform packages through GitHub Releases. The workflows below come from the browser collaboration demos. Feishu outputs require the relevant skills and authorization.
The browser collaboration preview brings Siada into the page you are reading. Your current reading, browsing history, and recent actions provide context for understanding, organizing, and creating.
Understand your intent. Find previously visited papers, GitHub projects, and articles, then compare them with the current page. Read recorded clicks, scrolling, and navigation to answer questions about what you just did.
Make room to create. Place translations, explanations, and comparisons beside the relevant paragraphs. Turn information from the web into reports, presentations, pages, or tools using the skills and integrations you have configured.
Browse efficiently. Search, navigate, and extract information across pages. Trace lets you inspect action snapshots, tool execution time, and intervals between calls to review the process.
| Workflow | Demonstrated result | Full video |
|---|---|---|
| Reading → Feishu sharing | Organize a daily article selection, retain source links, and share it with a study group. | ▶ 30 s |
| Browsing history → Paper comparison | Recover relevant research and compare it with five key insights in a survey, with explanations and comparisons placed in the original page. | ▶ 28 s |
| Recent actions → Browsing path | Summarize recorded clicks, scrolling, searches, and page navigation in a readable reply. | ▶ 28 s |
| Current page → Inline explanation | Put translations, annotations, and comparisons beside the corresponding paragraphs. | ▶ 28 s |
| Financial reports → Feishu report | Compare quarterly and annual revenue, gross margin, and cash flow, then organize analysis, charts, and sources in a document. | ▶ 32 s |
| Product research → Presentation | Collect product images and specifications, compare prices and features, and produce buying recommendations as Feishu slides. | ▶ 34 s |
| Browser search → Trace review | Find agent dependencies, versions, and sources in a GitHub project's pyproject.toml, then inspect the actions and their timing. |
▶ 36 s |
Click an animated preview to open the full MP4. The six demos are silent, with Chinese on-screen captions; the paper comparison demo also shows inline explanations.
| Reading → Feishu · 30 s | Browsing history → Paper · 28 s |
|---|---|
![]() |
![]() |
| Recent actions → Context · 28 s | Financial reports → Feishu · 32 s |
|---|---|
![]() |
![]() |
| Product research → Slides · 34 s | Browser search → Trace · 36 s |
|---|---|
![]() |
![]() |
See Browser collaboration preview for the integration overview. Follow Releases and PyPI for published versions.
| Capability | How it helps |
|---|---|
| Code development and debugging | Explore a repository, implement features, fix defects, and run relevant checks across languages and frameworks. |
| Long tasks and sub-agents | Plan complex work, delegate bounded tasks to specialized agents, and track progress toward a goal. |
| Model choice | Use different model families and configure custom API endpoints through the provider layer. |
| Persistent memory | Retain preferences, project context, and decisions across sessions, with a structured fact store for retrieval and feedback. |
| Skills, agents, and MCP | Add reusable workflows, specialized sub-agents, and tools from external MCP servers. |
| Session control | Resume conversations, compact context, and opt into checkpoints for recovery. |
| Multiple entry points | Work in the terminal, connect an ACP client, or send tasks through a configured Feishu/Lark bot. |
| Scheduled automation | Use the background daemon for recurring tasks and proactive workflows. |
| Requirement | Details |
|---|---|
| Operating system | macOS, Linux, or Windows |
| Python | 3.12–3.13; the installation command selects Python 3.12 |
| GCC | 11+ on macOS / Linux |
| Package manager | uv |
| Model access | A configured provider account or your own API key |
The terminal UI uses Node.js 24 LTS. Install Node.js yourself for terminal-only use, or use the runtime installed by siada-browser setup with a published browser platform package. The launcher does not download Node.js on startup. Source development and browser builds also need npm and Bun.
Install uv on macOS / Linux:
curl -LsSf https://astral.sh/uv/install.sh | shOr on Windows (PowerShell):
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"Install the latest stable release from PyPI with the same command on all platforms: The current stable release is 1.7.41.
uv tool install --force --python python3.12 --compile --with pip --default-index https://pypi.org/simple siada-cli@latest
uv tool update-shellOpen a new terminal if siada-cli is not yet on PATH.
Verify the installed version:
siada-cli --version
# siada-cli 1.7.41Launch Siada in the project you want to work on:
cd path/to/your/project
siada-cliComplete the provider, model, and API key setup when prompted, then describe a task:
Find the cause of the failing tests, fix it, and run the relevant test suite.
Use /model to select an available model. Configuration lives in ~/.siada-cli/conf.yaml; custom model definitions live in ~/.siada-cli/models.json. See the User Manual and External Model Configuration for details.
uv tool upgrade siada-cliuv tool uninstall siada-cliSub-agents retry up to twice when a response ends with reasoning and an empty final answer. Retries continue the existing session with completed tool results. If all three attempts return this result, the task is marked failed. This check detects a potentially incomplete response; it does not establish a network fault.
Open a running sub-agent's detail view to follow its reasoning as it streams. Each reasoning part updates in place and keeps the latest 5,000 characters for long responses. Providers that return only completed reasoning show it when the item arrives.
| Entry point | Usage |
|---|---|
| Interactive terminal | Run siada-cli for a continuous conversation in the terminal UI. |
| Script or CI task | Run siada-cli --prompt "Find and fix the failing tests" for a single task. |
| ACP client | Configure the client to launch siada-acp as its agent process. |
| Feishu / Lark | Configure a bot, then send tasks from chat; see the Remote Control Guide. |
| Browser preview | A development preview requiring the browser addon and Chrome extension; see the integration overview. |
Enter these commands inside an interactive session:
| Command | Purpose |
|---|---|
/help |
List available commands. |
/model |
Select or change the model. |
/resume |
Browse and resume saved sessions. |
/clear |
Start a fresh conversation context. |
/compact |
Compact the current conversation. |
/skill-list |
List available skills. |
/skill-reload |
Reload skills after adding or editing them. |
/goal <objective> |
Set a goal for the session; use /goal clear to clear it. |
Enable checkpoints when starting an interactive session:
siada-cli --checkpointingCheckpoints are opt-in and are unavailable in non-interactive mode. See the User Manual for restore, undo, and comparison commands.
The preview connects a Chrome extension to a local proxy and the Siada ACP agent. ACP handles the agent conversation; MCP exposes browser tools for reading pages, navigating, capturing snapshots, and adding inline annotations. Browser actions and page snapshots can be reviewed in Trace.
Browser platform packages contain Node.js, the proxy, its production dependencies, the web client, and the Chrome extension. Install the package matching your CLI version and platform with:
siada-browser setupLoad ~/.siada-cli/browser/extension manually through chrome://extensions. The extension discovers the local connection token automatically; remote connections require manual settings. Sign in from Chrome with API Key or Free Trial. After uv tool upgrade siada-cli, run siada-browser setup again and reload the extension in Chrome. See Browser installation for one-command installation on Linux/macOS and Windows, source installation, and release details.
Install Python 3.12 or 3.13, Poetry, and a maintained Node.js LTS release with npm. Node.js 24 LTS satisfies the current UI dependencies.
git clone https://github.com/liauto-siada/siada-cli.git
cd siada-cli
poetry install --with dev
npm --prefix siada_cli_ui ci
npm --prefix siada_cli_ui run build:all
poetry run siada-cliRebuild the UI after changing its source; the launcher uses the bundled UI. To run the standalone ACP agent from the checkout, use poetry run siada-acp.
| Directory | Responsibility |
|---|---|
siada/entrypoint/ |
CLI arguments and startup |
siada/acp_server/ |
ACP agent server |
siada/agent_hub/ |
Agents and orchestration |
siada/tools/, siada/services/ |
Tools and application services |
siada/provider/ |
Model provider integrations |
siada_cli_ui/ |
Terminal UI |
tests/, docs/ |
Tests and documentation |
| Document | Contents |
|---|---|
| User Manual | Configuration, usage modes, commands, checkpoints, and troubleshooting |
| Skills | Create and manage reusable workflows |
| Custom Agents | Define specialized sub-agents and their tools, models, and skills |
| External Model Configuration | Connect custom models and API endpoints |
| Feishu / Lark Remote Control | Configure the bot transport and chat access |
| Compaction Configuration | Configure conversation compression |
| Contributing | Development conventions and contribution workflow |
Bug fixes, features, documentation improvements, and reusable skills are welcome. Read the Contributing Guide for the contribution agreement and pull request requirements, and use GitHub Issues to report a problem or propose a change.
Thanks to the open source projects and contributors that make Siada possible, including:
- OpenAI Agents SDK — agent framework.
- LiteLLM — model provider integration.
- OpenHands ACI — agent tools.
- Gemini CLI — inspiration for terminal agent design.
- Agent Client Protocol — integration with agent clients.
See CREDITS.md for the full acknowledgements and license information.
Distributed under the Apache License 2.0.
See disclaimers.md.






