SoduWork
SoduWork is an open-source, AI-native enterprise collaboration platform. It connects people, AI agents, models, and business systems in request-centered workflows with acceptance criteria, AI review, human sign-off, and reusable plugins.
Visit mdiri-ai/SoduWorkOverview
SoduWork is an open-source, AI-native collaboration platform for enterprises. It connects people, AI agents, models, and business systems around shared requests, with acceptance criteria, independent AI review, human approval, and reusable plugins. The platform supports multiple agent harnesses, cloud or local models, and private deployment so teams can keep decisions, results, and proven methods within their own environment.
Key Features
- Multi-harness execution connects harnesses such as Pi, OpenCode, DeepSeek Harness, Qwen Code, Kimi Code, Codex, Claude Code, and Google Harness through one adapter interface, allowing users to switch mid-conversation while keeping history and the working folder.
- Separate implementation and review agents let one harness do the work and another review it against acceptance criteria, with a person making the final approval decision.
- Request-centered collaboration links requirements, execution, review, approvals, result versions, and evidence to the same task in a shared issue pool.
- Reusable plugins and skills turn proven chats and business methods into shared capabilities that can be reviewed, granted to teams or people, revoked, and reused across harnesses.
- Enterprise governance includes plugin review and permissions, roles, usage, cost, audit logs, and self-hosted deployment options.
- Model and runtime choice lets teams configure cloud models, local models, and execution resources for each stage, including long tasks sent to their own cloud runner.
Use Cases
- Engineering teams can move from requirements to release-ready software with code changes, test reports, cross-review, and release records tied to the same request.
- Business and operations teams can draft documents, create presentations, analyze spreadsheets and data, and turn meetings into minutes, decisions, and action items.
- Research and analysis teams can plan research, collect sources, cross-check evidence, and produce reports with citations and traceable sources.
- Teams can capture individual know-how as reviewed plugins so new members can follow proven methods and contribute improvements with appropriate permissions.
Getting Started
- Run from source with Python 3.12+, Node.js 22.19+, and Git installed.
- From the cloned repository, start the desktop app with sh start.sh --desktop on macOS or Linux, or .\start.cmd --desktop in Windows PowerShell.
- On first run, set the admin account password and add a model provider under Model catalog in the admin console.
- For evaluation, download the desktop client from Releases, enter the demo Hub address on the sign-in screen, then sign up and sign in.
- For Docker-based team deployment, run python3 ops/self-host/deploy.py up and then python3 ops/self-host/deploy.py credentials to show the initial admin password.
Deployment & Requirements
- Running from source requires Python 3.12+, Node.js 22.19+, and Git.
- The Docker team deployment starts the Hub, admin console, and PostgreSQL on one server.
- Docker must be installed and running for the Docker deployment path.
- An HTTPS reverse proxy should be placed in front of the Docker deployment to open it to a team.
- Private deployment lets organizations configure the environment around model, network, and permission requirements.
Before You Adopt
- License: MIT. Review its terms before using, modifying, or distributing the project.
- SoduWork is in public beta and should be evaluated before production use.
- The live demo is intended only for evaluation and should not receive real business or sensitive data.
- Plugin integrations depend on the APIs and permissions of existing business systems.
- A successful connection probe does not guarantee successful inference; model balance, permissions, and protocol support still matter.