User Manual Overview

This manual is for individual researchers, small teams, and admins. It focuses on the everyday flows in ResearchX: signing in, creating projects, starting conversations, managing files, using Assistants, installing capability packs, configuring models, and checking admin settings.

Recent Changes You May Notice

Conversations and Assistants

• Save reusable Assistant setups: Save a project's common model, instructions, memory, skills, actions, and runtime setup as an Assistant, then reuse it in another project or conversation.
• Use an Assistant only in the current chat: Try a different Assistant temporarily without changing the project default.
• Plan mode: Switch a single conversation to Plan so the Assistant first produces a reviewable implementation plan. After you confirm it, ResearchX exits planning and starts the implementation turn.
• Structured follow-up questions: When the Assistant needs your decision before continuing, the chat area shows a question card. You can select an option, provide a custom answer, or dismiss the request.
• Edit old messages and compare branches: When you edit a previous user message, ResearchX keeps the old conversation and creates a new branch, so you can compare different prompts or analysis paths.
• Long outputs are easier to read: Command results, file edits, and file writes are grouped into collapsible sections so the chat stays readable.
• File attachments are easier to add: Type @ in the composer to search project files and attach them to the message, or insert file references directly from the file panel.

Projects and Files

• More sidebar actions: Create projects, start chats across projects, and expand or collapse long conversation lists from the sidebar.
• Start the first conversation while creating a project: New projects can take you straight into the first chat.
• Delete projects from settings: Deleting a project also removes its conversations, files, and member links, so confirm carefully.
• Import files from another project: Choose a source project and copy selected files or folders into the current project.
• Safer uploads and renames: When uploaded files conflict, choose overwrite or automatic renaming. Rename operations now use an in-app input dialog instead of native browser prompts.
• Project storage quotas: When JuiceFS storage quota support is enabled, project lists and project settings can show used space and limits. Uploads that exceed quota are blocked with a clear message.
• Richer file preview: Common documents, spreadsheets, slides, images, and project-specific preview plugins can be viewed online.

Personal Settings and Models

• Language, theme, and interface choices are saved: Your preference stays consistent as you move through the app.
• More model provider templates: Model management includes templates for OpenAI Responses, OpenAI Codex, OpenRouter, DeepSeek, and other providers to reduce manual provider configuration.
• Codex OAuth model credentials: When using OpenAI Codex (ChatGPT OAuth), the model form can log in to Codex, import a local Codex login, refresh the token, or clear the saved credential.
• Personal model management: Users can manage personal models at /workspace/models; the chat model picker only shows models available to you.
• Token usage dashboard: Review model usage over the last 30 days by project, model, date, and other views.
• Billing and wallets: If billing is enabled, you can review wallet balance, credit limit, and usage history. New model calls may be blocked when balance is insufficient.
• Linux user binding: In Slurm deployments with Linux identity binding enabled, bind your Linux account from Profile so project owners can choose to run supported project work as that identity.

Skills, Tools, and Automation

• Install Skills more easily: Projects can install Skills from catalogs, zip imports, other projects, or global enablement.
• Import and export Actions and Agents: Reuse team workflows by packaging and moving them between projects.
• Approve risky tool actions before they run: When an Assistant wants to edit files or run a command that changes the environment, ResearchX can ask for your confirmation first.
• Review diffs and revert when possible: File changes can show a diff, and some completed changes can be reverted if the workspace has not changed since then.
• Scheduled tasks: Create recurring prompt tasks for daily summaries, weekly checks, or regular reports.
• Web search and external tools: When enabled by admins, Assistants can use web search or project-connected external tools.

Admin and Advanced Capabilities

• Public project showcase: Admins can publish selected projects to a read-only public showcase page.
• Custom page previews: HTML files and Visualizer plugins can open directly in the preview modal, which is useful for project visualization results.
• Clearer member and permission rules: Email invitations are validated, and project writing or container controls follow project roles.
• Restartable project containers: Project runtime environments can be started, stopped, retried, and configured with CPU, memory, or GPU resources. ResearchX also reconciles Docker / Kubernetes runtime state so the UI stays aligned with real containers.
• Grouped admin navigation: Admin settings are organized by organization and usage, AI capabilities, Agent policies, security and access, and runtime infrastructure.

See "Conversations and Messages", "Project Management", "File Management", and "Model Management" for detailed steps.

Assistant

An Assistant is a reusable way of working. It can remember the model, instructions, memory, skills, actions, and runtime environment that a project commonly uses, then apply that setup to a project or just one conversation.

You can open it from:

• the Assistant entry in the top-right chat settings menu
• the Assistant management page inside project settings
• admins can manage recommended platform templates from /workspace/admin/assistants

There are two common ways to use it:

• Apply to this project
  • New conversations in this project use this setup by default
  • Best for team projects, fixed research workflows, or long-running work
• Use in this chat
  • Only applies to the current conversation
  • Does not overwrite project settings
  • Best for trying another Assistant temporarily

Saved configs inside a project support these operations:

• Save current config: save the project's common settings
• Update config: replace an existing config with the current project settings
• Apply config: apply a saved config to the project
• Share / unshare: submit a config for admin review so it can become a team or platform template
• Delete config: remove saved configs that are no longer used

Admin template management supports:

• reviewing Assistants shared from projects
• pinning common templates and adjusting display order
• maintaining descriptions so users can quickly tell what a template is for
• controlling template visibility to avoid accidental use of stale or experimental configs

When an Assistant is enabled:

• The chat header shows the Assistant name
• Some project switches related to that Assistant are managed automatically, reducing conflicting settings
• If the Assistant needs a skill, action, or agent that is not installed in the current project, the page shows a warning so you can install it or choose another setup

Pinned Assistants in a new chat can be selected directly, and the More entry takes you to the full Assistant management page.

Channels

Channels connects external chat tools to ResearchX. Current project channels include Feishu and Weixin-style channels, which are useful when a team wants group chats, external threads, or bot messages to flow into a ResearchX project.

Think of it as a bridge that receives external messages into a project conversation and sends ResearchX replies back to the same external thread:

• Open it from the workspace left navigation: Channels
• Select a project first, then configure that project's channel connection
• Each project can keep its own external channel configuration
• Incoming external chats are bound to ResearchX conversations so later replies return to the same external thread

Recommended setup flow:

1. Go to /workspace/channels
2. Select the project to connect
3. Enter the credentials required by the channel, such as Feishu App ID and App secret
4. In Feishu Open Platform event settings, choose long connection delivery and subscribe to im.message.receive_v1
5. Save and use Test connection to verify credentials
6. Click Connect or Reconnect to establish the WebSocket connection
7. Send a real external message and confirm that an active binding appears on the page

Message behavior:

• Feishu inbound events can acknowledge user messages quickly with a reaction-style signal while ResearchX starts processing.
• Weixin channels try to show a typing/activity indicator while the Assistant is generating a reply, then clear that state when the final message is sent.
• In shared_bot mode, external channels can use built-in tools and configured project capabilities. File-changing or high-risk work still follows the project's confirmation and permission rules.

MCP (Model Context Protocol) Servers

An MCP server is a way to connect external tools to the current project. After it is configured, the Assistant can call those tools during a conversation.

The project assets modal includes an MCP tab for managing these external tool services:

• Click New and enter a name plus service URL
• Use Test & Load Tools to check the connection and load available tools
• Save and enable the service so its tools can appear in conversations
• Use Expose tools to chat to decide whether the Assistant can see those tools
• For higher-risk tools, require user confirmation before the Assistant can call them; this follows the normal tool permission confirmation flow

Recommended Reading Order

1. Quick Start
2. Core Concepts
3. Account and Authentication
4. Project Management
5. Container Management
6. Agent / Action / Runtime / Visualizer Management
7. Conversations and Messages
8. Web Search
9. MCP Servers
10. Skill Management
11. File Management
12. Model Management (Owner)
13. Model Billing and Wallets
14. Collaboration and Permission Boundaries
15. Container Mounts (Admin)
16. Retrieval Memory
17. Resource Quotas
18. Execution Platform
19. FAQ
20. Troubleshooting and Support

Scope

• This manual prioritizes shipped features and standard user flows.
• Planned capabilities are explicitly marked as future versions.