Documentation

Installation & Setup

Download Installer

Tokeny supports Windows, macOS, and Linux. Go to the download section on the homepage — it automatically detects your device and recommends the right installer, and you can also switch platforms manually:

• Windows: Download the .exe installer and double-click it to run the setup wizard
• macOS: Download the .dmg file and drag it to the Applications folder (separate builds for Apple silicon and Intel chips)
• Linux: Download the Linux (x64) package

First Launch & Sign-in

Tokeny is ready to use right after the first launch. The app creates a local SQLite database to store your conversations, settings, and file data.

After launching, you can sign in with your Tokeny unified account — the same account used on the official website tokeny-ai.com. Sign in with your phone number (SMS code or password) or by scanning a WeChat QR code. Signing in unlocks online features such as membership benefits and one-click installs and submissions in the marketplace. Without signing in, you can still use local features with your own provider API Keys.

System Requirements

Built-in vs System Environment

Tokeny offers two ways to manage runtime environments, ensuring you can get started without complex configuration:

Built-in environment initialization steps:

1. 1Open the app and navigate to "MCP → Environment Check"
2. 2If you see Node.js, Python, or Git showing "Not Extracted" status, click the "Initialize Built-in Environment" button
3. 3Wait for the extraction to complete (usually under a minute). A smoke test will run automatically after extraction
4. 4All items showing green checkmarks means the environment is ready

Configure AI Model

Supported Providers

Tokeny comes with 10 built-in model channels, and also supports local deployment and custom channels:

In addition, Ollama (local deployment) and custom channels are supported (OpenAI / Anthropic protocol compatible, with custom Base URL). The model list is continuously updated with each release — refer to the app for the latest.

Configure API Key

1. 1Click "AI Model" in the left navigation bar to enter the model configuration page
2. 2Select the provider you want to use
3. 3Enter your API Key in the corresponding provider section
4. 4Click "Test Connection" to verify the Key is valid

Select & Switch Models

After configuring an API Key, you can select a default model from the model list. You can also switch between different providers and models during a conversation — new messages will use the newly selected model. Each session can have its own model selection, and multiple sessions within the same workspace don't interfere with each other.

Dual Protocol Switching

Some providers offer both OpenAI-compatible and Anthropic-compatible API protocols. You can switch between protocols with one click on the model configuration page, and the Base URL will update automatically.

Currently supported providers for dual protocol switching:

Token Plan

Some providers offer Token Plan subscription billing (monthly/quota-based). When enabled, API calls for that provider are no longer billed per token. Tokeny automatically switches to the plan-specific API endpoint and sets cost tracking to zero.

Currently supported providers and applicable models:

To enable: go to the "AI Model" settings page and toggle on Token Plan in the corresponding provider section. Once enabled:

• Base URL automatically switches to the plan-specific endpoint; restored when disabled
• Models outside the plan are automatically disabled; restored when disabled
• Cost tracking for that provider shows zero (token usage is still recorded)

Recommended Models

Tokeny is an AI Agent application — the AI doesn't just "chat" but also "takes action": calling tools, reading/writing files, executing commands, multi-step planning, and decision-making. This requires models with strong Agent capabilities (function calling / tool use / instruction following), not just conversational ability.

Below are our recommendations based on real-world usage (pricing in CNY per million tokens):

High Performance

Top-tier Agent capabilities, ideal for complex multi-step tasks and professional scenarios

Best Value

Extremely low cost or free, excellent for everyday use and lightweight tasks

🧠 Thinking Mode   🔧 Tool Calling   👁️ Vision   Pricing format: Input / Output (CNY per million tokens). Models and pricing may change as providers adjust them — refer to the in-app model list and each provider's official website.

Not Recommended for Agent

The following types of models perform well in specific scenarios but are not suitable as primary models for Tokeny, as they have significant shortcomings in tool calling and Agent tasks:

Thinking Mode

Some recommended models (such as DeepSeek-V3.2, Kimi K2.5, Qwen 3.5 Plus, etc.) support a toggleable "Thinking Mode." When enabled, the model performs a round of reasoning before answering, which is ideal for complex questions requiring deep analysis. Unlike pure reasoning models like R1, these models retain full tool-calling capabilities in thinking mode and do not affect Agent task execution.

For custom models, you can manually declare whether they support thinking. The thinking mode toggle is remembered per workspace, so each workspace keeps its own setting.

Chat Features

Basic Chat

Type your question or task description in the input box at the bottom of the main interface and press Enter to send. The AI will understand your intent based on context and provide a response. Multi-turn conversations are supported — the AI automatically remembers all context within the current session. You can also click the microphone button next to the input box for voice input.

Voice Input

Click the microphone button next to the input box and start speaking — the system automatically transcribes your speech into text with smart sentence segmentation. Chinese speech is polished before it lands in the input box: filler words are removed and punctuation is tidied up. Review the text, then send it.

Plan Mode

For multi-step or high-risk tasks, you can enable Plan Mode: the AI first presents an action plan, and only executes it step by step after your review and approval. You can pause execution at any time and approve each step individually. Ideal for complex tasks where you need tight control over the execution process.

Session Management

Sessions are organized by workspace. You can create multiple independent sessions within the same workspace — chat history is fully isolated, and each session can have its own model selection without affecting the others.

• The session switcher is located in the title bar — click to switch quickly
• Full-text search across conversation history: enter keywords to quickly find content in past sessions
• Unsent drafts and AI responses are saved automatically — nothing is lost even if the app quits unexpectedly

Files & Images

You can send files and images in conversations. Drag files into the input box or click the attachment button to select files. The AI can understand text files, code files, images, and many other formats, analyzing and answering based on file content.

Streaming Output

All AI responses use streaming output, so you can watch the generation process in real time. The AI's thinking process can be expanded or collapsed, displayed with a typewriter effect. Failed API calls are retried automatically, so network fluctuations won't interrupt your conversation. If you're not satisfied with the current response, you can click the "Stop" button to interrupt the output.

Message Queue

While the AI is executing a task (such as running commands, calling tools, or generating reports), you can continue typing new messages. These messages won't be lost — they automatically enter the message queue and wait for the AI to finish the current task before being processed in order.

• Messages sent while AI is busy will show a "Queued" tag
• After the AI completes its current task, it automatically picks up the next message from the queue
• Queued messages can be canceled at any time (click the cancel button next to the message)
• Multiple messages are processed in the order they were sent (first in, first out)

Interrupt & Priority

If the AI is executing a task you don't want, or you have a more urgent need, you can interrupt the current task at any time:

• Stop Generation: Click the "Stop" button, and the AI will immediately halt the current task. Any content already generated will be preserved in the conversation and won't be lost.
• Send Immediately (Priority): Queued messages support a "Send Immediately" action. Clicking it will interrupt the AI's current task and immediately start processing this priority message.

Skill System

What is a Skill

Skills are Tokeny's core capability extension mechanism. Each Skill is an independent professional module defined through a standardized SKILL.md file that specifies its capabilities, trigger conditions, and behavioral instructions. When a Skill is activated, its professional instructions are injected into the AI's prompt, giving the AI specialized expertise in that domain.

Built-in Skills

Tokeny ships with multiple professional built-in skills, ready to use out of the box. The commonly used ones are listed below:

Manual & Auto Routing

Manual trigger: Type / in the input box to see all available skills and select one to trigger. For example, typing /deep-research current state of quantum computing will launch the deep research skill.

Auto routing: The AI automatically determines whether to activate a skill based on your conversation content. For example, when you say "help me search for the latest AI news," the system will automatically identify and activate the web search skill.

Skill Marketplace

The Tokeny Skill Marketplace features official and community skills. Visit tokeny-ai.com/skills to browse featured, popular, and latest sections, filter by category, or search; you can also browse the marketplace inside the client and install skills with one click.

On a skill's detail page, click "Install to Tokeny" to launch the client and install it in one step, or download the ZIP package and import it manually.

Official extension examples:

• Word Document Processing: Read and generate Word documents
• Excel Spreadsheet Processing: Read and generate Excel spreadsheets
• PDF Tools: PDF file parsing and processing

Custom Skills

Beyond built-in and marketplace skills, you can create your own Skills:

• Define capabilities with a SKILL.md file — supports file editing and hot reload, with changes taking effect immediately
• Install from the marketplace, a GitHub repository (including monorepo subpaths), or a local workspace
• Skills can have a Chinese display name, which is written into SKILL.md on installation and shown in the "Installed" list

The AI also suggests turning frequently used operations into skills — use /save-as-skill to save the current workflow as a reusable Skill.

Submit & Publish

Click "Submit" on the marketplace page to open the submission wizard:

1. 1Import — provide a GitHub link, paste the full SKILL.md content, or upload a ZIP package
2. 2Preview — the name, summary, and documentation are parsed automatically
3. 3Confirm & submit — complete the category and tags; for the first release, upload the full ZIP package (up to 30 MB) and provide a version number

Submission limits are 5 per hour and 10 per day. Each submission goes through one review: once approved it is published immediately; if rejected, you can revise and resubmit. Manage all submissions and versions under "Profile → My Submissions" on the website.

Skill Management

Navigate to the "Skills" page from the left navigation bar to view all installed skills. Click any skill to see its details, including the SKILL.md source content. Here you can install, uninstall, or review the full prompt of third-party skills.

MCP Tool Extensions

What is MCP

MCP (Model Context Protocol) is an open tool extension protocol that allows the AI to call external services to accomplish specific tasks. Through MCP, Tokeny can connect to various third-party services such as database queries, email sending, browser automation, and more, greatly expanding the AI's practical capabilities.

Built-in Services

Tokeny includes the following built-in MCP services, ready to use without additional configuration:

• WebFetch: Retrieve web page content, supports HTML-to-Markdown conversion with built-in SSRF protection
• Web Search: Get the latest information from the internet via search engines
• Playwright Browser: Automated browser operations with three connection modes — launch a new instance, reuse your local browser Profile, or connect directly via CDP; tool connections are established asynchronously in the background, so a dropped service connection never stalls AI replies
• Email Service: Supports IMAP receiving and SMTP sending, configurable multiple email accounts for automated email processing
• Tokeny Connector Browser Extension: Takes over your already logged-in browser to auto-fill forms, scrape data, and submit, with screenshots captured throughout as evidence

MCP Marketplace

The official MCP Marketplace (tokeny-ai.com/mcp) features official and community MCP Servers. Browse by category or search; each detail page provides the install command and remote URL with one-click copy, and you can also launch the client to install with one click.

You are welcome to submit your own MCP Server — import from a GitHub repository or connect via a remote URL. The submission entry is the "Submit" button on the marketplace page.

Add Third-Party Services

1. 1Go to "Settings → MCP Services"
2. 2Click the "Add Service" button
3. 3Choose the connection method: SSE (network URL), stdio (local command), or StreamableHTTP
4. 4Enter the service URL or startup command
5. 5After saving, the system automatically connects and loads the tool list

Service Management

On the MCP management page, you can view the status of all connected services (connecting/connected/disconnected), available tool lists, and runtime logs. You can manually reconnect, disable, or delete services. The system has a built-in automatic reconnection mechanism that restores connections after network recovery.

Built-in Tools

Tokeny includes 30+ built-in tools covering file operations, command execution, search, media generation, automation, and more. You can manage the enable/disable status of each tool on the "Tools" page within the app.

File Operations

The AI can perform file operations within your workspace. 10 operations are supported:

Shell Command Execution

The AI can generate and execute Shell commands to accomplish tasks like installing dependencies, running scripts, and building projects. Real-time output and interactive commands are supported. An approval dialog appears before each command execution, giving you the following options:

• Approve: Approve only this specific command. After execution, if the AI needs to run another command, a new approval dialog will appear. This is the safest approach, recommended for everyday use.
• Auto-execute: Once enabled, all subsequent Shell commands generated by the AI in the current session will execute automatically without individual approval dialogs. Suitable when you trust the current task and need to run many commands consecutively (e.g., project initialization, bulk dependency installation).
• Reject: Cancel execution of the current command. The AI will adjust its approach based on your feedback.

Search & Locate

The AI can quickly locate files and content within the workspace:

• grep — Content search, supports regular expressions, file type filtering, and recursive search
• glob — Path matching, quickly locate files by filename patterns

User Interaction & Planning

The AI proactively collaborates with you when executing complex tasks:

• Ask User — Proactively asks you questions when additional information is needed
• Task Planning — Breaks complex tasks into step-by-step checklists with real-time progress updates
• Progress Reports — Reports current progress during long-running tasks
• File Preview — Preview generated files directly within the app

Media Generation

The AI can call configured media models to generate images and videos. You can specify separate models for image generation and video generation in the "AI Model" settings (e.g., SiliconFlow, Doubao, etc.).

Multi-Agent Collaboration

Tokeny supports multiple forms of multi-agent collaboration:

• Group Chat Collaboration — Multiple agents collaborate in the same conversation, handing off via @ mentions to complete tasks together
• Sub-task Delegation — Delegate sub-tasks to independent Agent instances for isolated execution, avoiding context interference; results are reported back to the main task upon completion
• Preset Agents — 25 built-in roles (Tech Lead, Deep Researcher, Meeting Minutes, Bookkeeping Assistant, and more), ready to use out of the box
• Custom Agents — Create your own agent with a single natural-language description

The AI can also create and manage Cron scheduled tasks — see the "Scheduled Tasks" chapter for details.

Management & Extensions

The AI can manage Skills and MCP services directly within the conversation, without needing to navigate to the settings page:

• Skill Management — List, install, uninstall Skills (supports installation from URL, marketplace, or workspace)
• MCP Management — List, add, remove, start/stop MCP services and query available tools
• Security Audit — Perform AI semantic security analysis on high-risk commands
• Terminal Preference — Specify your preferred command-line tool for smarter detection at runtime

Safety & Approval

Approval Workflow

When the AI needs to perform high-risk operations (such as running Shell commands or deleting files), the system automatically pauses and displays an approval dialog. You have three options:

• Approve: Confirm and execute the original command
• Edit & Execute: Modify the command content before execution
• Reject: Cancel execution; the AI will adjust its approach based on your feedback

Dual-Layer Security Review

Tokeny employs a dual-layer security review architecture to ensure thorough evaluation of high-risk operations:

• Layer 1: Rapid Risk Identification: The system automatically assesses risk for each command, identifies dangerous operation patterns, and labels risk levels with extremely fast response.
• Layer 2: AI Deep Analysis: For commands flagged as high-risk, additional AI semantic analysis is performed to generate a risk summary and detailed explanation, displayed in the approval dialog for your reference.

Permission Control

Tokeny provides a three-tier permission control system, configured independently per workspace to meet different security needs:

In addition, the system can detect approval deadlocks and proactively alert you, so tasks never get stuck in repeated confirmation loops.

File Protection

The file system security validator provides three-stage path verification: path normalization, path traversal attack detection, and symlink target verification. All file operations are restricted to the workspace sandbox, ensuring the AI cannot access or modify critical system files.

Workspace

File Management

The workspace is a shared file directory between you and the AI. You can browse the file tree, create folders, and upload files from the right panel. Files generated by the AI (such as research reports, code files, and PPTs) are all saved in the workspace. File management also includes the following capabilities:

• The file tree watches external changes in real time and refreshes automatically
• Right-click context menu operations and inline renaming are supported
• External editors can be detected and launched to open files

File Preview

The workspace includes a built-in multi-format file previewer supporting the following formats:

• Markdown (.md): Rendered rich-text display
• Word Documents (.docx): Full document preview
• PPT Presentations: Reveal.js slideshow preview and playback
• HTML Projects: Embedded browser rendering with live preview for multi-file projects
• Code Files: Syntax-highlighted display
• Image Files: Preview for common image formats
• Video Files: Built-in player preview

Report Export

Reports and documents in the workspace can be exported as PDF or Markdown formats. Click the export button on the file preview interface and select the target format. PDF export uses a built-in rendering engine to ensure styles match the preview.

Smart Memory

Automatic Memory Extraction

After a conversation ends, Tokeny asynchronously extracts your preferences and key facts. The memory system uses a dual-layer architecture with structured categorization, so the AI understands you better over time:

• Dual-Layer Memory Architecture: Global memories take effect across workspaces, while workspace memories apply only to the current project.
• Seven Structured Categories: Memories are archived into seven categories: preferences, facts, instructions, events, profiles, relationships, and goals.
• Three-Stage Smart Injection: In your next conversation, relevant memories enter the AI context through a three-stage smart injection mechanism.
• Full-Text Search & Conflict Merging: Memories support full-text search, and conflicting memories are merged automatically.

Memory Maintenance

A background maintenance system periodically organizes your memories to keep the memory store healthy:

• Clean up expired memories
• Decay memories that have been inactive for a long time
• Merge fragmented memories
• Detect and flag contradictory memories

You can assign a dedicated model to the memory system (for example, a cheaper model to handle memory extraction and maintenance) without affecting your main conversation model.

Memory Management

Navigate to the "Memory" page from the left navigation bar to view, edit, or delete saved memory entries. Memories are displayed as structured cards with category filtering and full-text search. You can manually add important personal preferences or remove inaccurate memories.

Scheduled Tasks

Create a Task

You can create automatically executing tasks in "Settings → Scheduled Tasks." Each task includes a natural language instruction and an execution schedule. When the specified time arrives, the system automatically creates a new session and executes the task instruction.

Cron Expressions

Scheduled tasks use Cron expressions to define execution times. Common examples:

Task Management

Scheduled tasks have a dedicated overview page that brings together the status and last run time of every task. Task results are presented as notification cards, so you can follow the execution in real time. You can enable/disable, edit, and delete tasks at any time.

Smart Assets

What Are Smart Assets

Reports, presentations, notes, images and other deliverables produced by the AI during conversations are automatically collected into the local Smart Assets library (Labs) for unified management — no longer scattered across different places.

Browsing & Management

The asset library supports tag-cloud filtering, preview popups, and full create/read/update/delete operations. With built-in full-text search, you can locate any past deliverable locally in seconds.

Referencing in Tasks

During conversations, you can reference content from the asset library as source material, letting the AI build on your existing work to keep creating.

IM Integration

What is IM Integration

IM integration allows you to chat with Tokeny directly through enterprise instant messaging platforms (such as Feishu/Lark). Once configured, messages sent in the IM are automatically received and processed by Tokeny, with responses sent back to the IM chat window. You can use AI capabilities remotely without opening the desktop application.

Supported Platforms

Currently supported IM platforms:

• Feishu/Lark — Full support: direct execution mode with no repeated confirmation cards; the tool system covers two modules — messaging/group chat and platform capabilities — letting you work with Bitable, Docs, Calendar, Contacts, Drive, and Wiki; MCP external tools can also be invoked right from the IM
• WeCom (WeChat Work) — Interface reserved, coming soon
• DingTalk — Interface reserved, coming soon

Configuration & Usage

Using Feishu/Lark as an example, the setup steps are:

1. 1Create a custom app (bot type) in the Feishu/Lark developer console
2. 2Obtain the App ID and App Secret
3. 3Navigate to the "IM" page in Tokeny's left navigation bar and enter the credentials
4. 4After enabling the connection, Tokeny will automatically receive Feishu messages via long-polling
5. 5In Feishu, @ your bot or send a direct message to interact with Tokeny

Once connected, you can use all of Tokeny's core features directly in the IM: chat, trigger Skills, execute tools, etc. IM sessions and desktop app sessions are independent, each with their own context.

Theme & Settings

Theme Switching

Tokeny supports three theme modes: Dark, Light, and Follow System. Click the theme button at the bottom of the left navigation bar to cycle through them. When "Follow System" is selected, the app automatically matches your operating system's dark/light setting. Theme preferences are saved automatically.

Mini Mode

Mini Mode is a compact window form that keeps only the core chat interface — perfect for pinning to a corner of your desktop for quick questions anytime. You can switch back to the full window at any time.

Data Storage

The desktop app can store the following in its local database or workspace directories. Accounts, online models and enabled services may also involve server processing:

• Conversation history and message records
• Application settings and preferences
• Workspace files and directories
• Persistent memory data
• API Keys (encrypted storage)
• Scheduled task configurations and execution records

Token Usage Statistics

In the "Settings" panel, you can view detailed Token usage statistics to help you understand and control AI usage costs:

• View input/output Token consumption per session
• Breakdown by model and provider
• Real-time display of current conversation Token cost
• Sub-task and background task usage also tracked

The Settings page is organized into tabs by topic (Models, Tools, Scheduled Tasks, and more), keeping information clearly structured.

Keyboard Shortcuts