Tokeny FAQ: models, costs, offline use and files

What is Tokeny, and what work is it for?

Tokeny is a desktop AI assistant operated by 北京星河千帆科技有限公司. Its official website is tokeny-ai.com. It connects models, workspace files and tools for research organization, documents, spreadsheets, presentations and development tasks on Windows, macOS and Linux. Describe a goal in natural language, then inspect the actual files saved to your workspace.

For tasks that read or modify files and call tools, start with meeting notes to Word or CSV to Excel. For a short question, use the chat tool that suits you. Check each tool's file access, permissions and extensions separately.

Can I use it without programming?

Describe everyday tasks in natural language, such as “Turn these meeting notes into minutes with owners and open questions.” First connect an available model and select the workspace containing your input. API keys, custom models and MCP servers may require endpoints, credentials or runtimes; installing Tokeny does not configure every third-party tool. See your first task.

Why is there no first-launch setup wizard?

Tokeny 1.7.8 removed the initial setup wizard. Accept the terms and privacy policy to enter the main workspace while bundled runtimes prepare in the background. If no model is configured, use Configure AI model beside the composer. Create a workspace from the New menu at the top and configure other options as needed.

How do I connect DeepSeek, Qwen or Doubao models?

Open Settings → AI Models, select the provider channel, enter its API key and enable a model. Custom channels also require the correct API URL, protocol and model ID. Save, test the connection and check the model in your conversation. Confirm whether a web chat subscription includes API access with the provider; a chat account is not an API key. See models and data flows.

Do I need a Tokeny account?

You can use your own configured provider API key for local workspace tasks without signing in to Tokeny. Built-in channels, membership and account-based marketplace actions require the unified Tokeny account and applicable entitlements. Sign-in status does not determine whether inference is online: an API key for an online provider still sends requests to that provider.

How are model usage and membership billed?

Your provider bills calls made with your own API key. Tokeny's built-in channels use the models, quota and entitlements available to your account. Check current account and provider pages for availability and pricing. Parallel subagents, memory organization and context summaries may add model calls; inspect usage with a small task first. The membership page shows the account entry point.

Can it work entirely offline?

You can manage locally stored history and files on your computer. Generating new answers offline requires an available local inference service and local access to every tool and source used by the task. Online models, web search, remote MCP servers and downloads still require a network connection. Selecting a local model does not make every part of a task offline.

Is local file content sent to a server?

Local storage does not mean every operation runs on your device. Online models may receive prompts, selected context, file excerpts and tool results. Accounts and other online features can also involve servers. Before using business documents, check model endpoints and tool scope, and read data flows and the privacy policy.

How do skills and MCP differ? Do I need both?

A skill describes a task's method, steps and output requirements. MCP exposes callable tools. Use them independently or together, for example extracting input with MarkItDown MCP and preparing a Word report with the docx skill. Required extensions depend on the task, and a successful connection still needs task-level verification. See skills vs MCP.

Can I deliver a generated Excel or Word file immediately?

Open and review it first. For Excel, check input rows, data types, formula references and recalculated totals. For Word, check facts, owners, dates and unresolved questions. The first workbook in our CSV tutorial required formula corrections before it passed Excel recalculation. Download the tested outputs and review method to apply the same checks to your work.

How do I stop multiple running subtasks?

Inspect prerequisites, dependent tasks and status in the Agent relationship graph. Use Stop in the composer and wait for execution to end before reviewing files. Stopping does not undo completed writes. Exiting or restarting does not automatically replay unfinished dependency chains; identify completed work and remaining steps before continuing. See subagent collaboration.

The Tokeny team maintains this page, with product workflows checked against version 1.7.8. Consult the current documentation and release history for changes, or report discrepancies through official contact.