TabTin - An open-source, full-stack AI agent team collaboration platform

What is TabTin?

TabTin is a full-stack, open-source AI agent team collaboration platform. The platform truly integrates agents into team workflows; tasks can be initiated on desktops, viewed on mobile devices, and executed by server agents, with complete context handover to colleagues. TabTin enables real-time collaboration between humans and agents within the same document, while deliverables require human verification, addressing the pain point of agents writing code quickly but teams repeatedly struggling to understand the context.

TabTin's main functions

• intelligent agent runtimeIt can orchestrate complex tasks, support cross-platform execution and permission control, require approval for high-risk operations, and allow rollback of version checkpoints.
• Collaborative DocumentsHumans and agents can collaborate in real time on the same document/spreadsheet, automatically merging and editing concurrently based on CRDT, without the need to export or switch tools.
• Task handoverThe task can be handed over to colleagues with the complete conversation history, documents, and attachments, and can be continued independently from any node by forking.
• project managementUse Issues to organize long-term collaborations, designate responsible persons, associate deliverables, and require manual verification of Agent results.
• AI Roles and SkillsSkills creates independent AI roles for different workflows and isolates memories, encapsulating repetitive processes into organization-level reusable components.
• Cross-device synchronizationThe desktop version retains the complete context, while the mobile version allows users to view progress, approve operations, and remotely control tasks. The Daemon executes long-running tasks.
• Organizational ManagementUnified management model access and budget quotas, and statistics on token consumption by member, Persona, and task dimensions.

TabTin's technical principles

• Runtime ArchitectureIt adopts a unified Harness to integrate organizational context, Skills, toolchain and execution environment. Agents dynamically decompose tasks and delegate to sub-Agents. Permission pipelines intercept high-risk operations. Version checkpoints record status to support recovery or rollback.
• Real-time collaboration engineTabDoc and TabData share a CRDT collaboration layer implemented in Yjs, Collab Live enables real-time merging and concurrent editing, and the Django backend handles access control, persistent storage, and version history in a unified manner.
• Context inheritance mechanismThe task object encapsulates the dialogue history, tool call output, and shared resource references. The visibility is controlled by the sharing level. Fork creates a branch from any node in the task graph and copies the context to the new workspace. The original task and the branch evolve independently.
• Workflow state machineUsing Issues as the unit of collaboration, Django persists metadata, assigns responsible persons, and associates resources. Agent task nodes and deliverables are linked via foreign keys, and status transitions require approval from a designated Owner or Reviewer.
• Role Isolation SystemThe Persona layer isolates system prompts, model configurations, and memory vectors for each role. Skills encapsulate tool call sequences and inspection rules with declarative configuration, and combine them with a timer to achieve periodic scanning and automated triggering.
• Multi-device synchronization protocolThe desktop client maintains the file system and terminal session state through a local daemon process. The mobile client establishes a temporary security control channel after authorization, and sensitive operations are confirmed twice. The daemon asynchronously executes long-running tasks and pushes state changes to all clients via WebSocket.

Follow and reply on WeChat“open source",join inAI open source project discussion group

How to use TabTin

• Visit the official websiteVisit the TabTin official website https://tabtin.com/ to browse the product architecture and feature introduction, and click the download link to obtain the desktop installation package.
• Install desktop clientOnce installed and logged into your account on Mac or Windows, the desktop version will retain the complete local working context and terminal environment.
• Import code repositoryBy connecting a real project repository to TabTin, the Agent can directly read code, run terminal commands, and operate the browser.
• Initiate the first taskDescribe your requirements in the task panel. The agent will automatically break down the target, call the Skills, and execute them. High-risk operations will be paused and await your approval.
• Real-time collaborative editingEdit documents and tables in TabDoc or TabData simultaneously with the Agent, merge concurrent modifications in real time, and automatically save version history.
• Task handover to colleaguesSend the task card along with the complete conversation history, shared documents, and attachments to team members. They can view the task according to their permissions or fork it to take over and continue the process.
• Project-based long-term trackingCreate a Project and organize collaborations for several weeks or more using Issues, assign an Owner for acceptance, and associate Agent tasks, code changes, and documentation with the corresponding Issues.
• Customized AI WorkflowCreate independent AI Personas for scenarios such as coding and PR review, configure dedicated models and isolated memories, and package repetitive processes into Skills for the team to reuse.
• Cross-device on-the-go controlUsers can check progress, reply to the Agent, and approve operations via mobile phone or webpage. Long-running tasks are handed over to the Daemon background for execution, and the results are notified across all platforms.

TabTin's core advantages

• Full-stack open sourceDesktop, mobile, server, and Agent Runtime are all open, allowing teams to deploy and customize them independently.
• Lossless context handoverThe task involves circulating a complete dialogue history, documents, and attachments among members to avoid repeated explanations of the background and repeated consumption of tokens.
• Human-computer collaboration on the same screenHumans and agents can edit in real time within the same document and form, with automatic merging based on CRDT, eliminating the need for exporting or switching tools.
• Delivery is controllable and acceptable.The Agent's execution result is only considered complete after approval by the designated Owner or Reviewer, ensuring that the code is deliverable rather than something no one dares to use.
• Continuous state across all terminalsTasks are initiated from the desktop, viewed on the mobile phone, and executed by the server agent. Long-running tasks are handled by the Daemon in the background and are synchronized across all devices.
• Organizational-level budget controlUnified management model access and token budget, accurately track consumption by member, AI role and task dimensions to avoid cost out-of-control.

TabTin's project address

• Project official website：https://tabtin.com/
• GitHub repository：https://github.com/tabtin-ai/TabTin

TabTin's Competitive Product Comparison

Application scenarios of TabTin

• Requirements gathering and documentationThe Agent researches competitors and generates a TabDoc, which the Product Manager then modifies and confirms directly within the same document before handing it over to the R&D team for further development.
• Multi-person relay developmentAfter the front-end finishes writing the component, it forks the task to the back-end, which automatically inherits the interface definition and dialogue history without needing to re-align the background.
• Asynchronous collaboration across time zones: Hand over long-running build or data crawling tasks to Daemon at night, and notify all platforms of the results the next day. Overseas colleagues can directly accept the results.
• Standardized code reviewCreate a dedicated AI Persona for PR reviews, and encapsulate the team's review rules as Skills to ensure consistent standards for each review.
• Getting Started with New ProjectsBy linking historical research, technical solutions, and issues to the Project, new members can quickly understand the project's context by reviewing the Agent's execution process.