1. Overview: what JarvisQwen is
JarvisQwen is an AI research assistant that runs autonomously, 24/7, in the cloud. It is designed for work that needs continuous tracking rather than one-off answers: after you submit a task or subscribe to a topic once, the system automatically searches, filters, archives and summarizes new information on schedule, then organizes the results into a searchable library and a daily briefing. The system keeps running while your browser is closed and your computer is off.
The table below summarizes how JarvisQwen differs from conversational AI assistants (such as Doubao or DeepSeek):
| Conversational AI | JarvisQwen | |
|---|---|---|
| How work is initiated | Question-and-answer; stops when the conversation ends | Runs continuously on schedule, without manual prompting |
| User presence | The user must stay online and wait | The user can go offline; execution continues in the cloud |
| Where results are kept | Scattered across chat history | Archived in the Library and digested into daily Briefings |
| Memory of the user | Limited memory across conversations | Long-term memory of your research focus and preferences |
| Cost management | Subscription or per-call billing, with little visibility | Hard daily budget cap plus a per-call audit trail |
2. Quick start
- Configure the system (Settings). Paste an API key, set a daily budget cap, and describe your research focus. If no key is configured, the system runs in dry-run mode: the full pipeline works with simulated model responses, at zero cost.
- Submit a task (Tasks). Describe what you want in natural language β for example, "survey the latest research progress on topic X" β and submit. The task detail page shows each execution step advancing in real time; you may close the page at any point.
- View the results (task detail page). When the task is marked Done, open its detail page and read the full output in the "Artifacts" section. The same results are also archived into the Library and included in the next day's Briefing.
Going further: add long-term topics in Subscriptions, and the system will track them automatically β briefings will arrive every morning without any further action.
3. Feature pages, in sidebar order
This page. It introduces what the system does, how to get started, and what each page is for.
The system overview: today's spending against the budget, the number of running tasks, and recent model-call statistics. A recommended starting point for each visit.
The core working page. Submit tasks in natural language, monitor their execution progress, and open a task's detail page to read its results in the Artifacts section.
Manages the topics you track long-term. Once a subscription is added, the system searches for new content on schedule β no need to submit the same task repeatedly.
The archive of everything the system has collected and summarized, with search. Unlike chat history, archived content is never buried or lost.
The daily digest the system generates automatically each morning, condensing everything newly collected in the past day.
The confirmation queue for high-risk operations such as deletion or outbound sending. Such operations are held here and executed only after your explicit approval.
The itemized record of every model call: which model was used, how many tokens were consumed, and how much it cost.
Links a Telegram account, so you can submit tasks and receive briefings from your phone.
A step-by-step guide to deploying your own private instance of JarvisQwen.
Global configuration: API keys, the daily budget cap, your research focus, and other preferences.
The technical reference: cost architecture, safety mechanisms, glossary, and troubleshooting, in greater depth than this guide.
4. Core concepts
Each task is broken into sequential steps (for example: search β de-duplicate β archive β summarize β memorize). The diagram at the top of the task detail page shows the real-time status of each step; a green node means that step has completed. Note that the diagram shows the execution process β the actual results are in the Artifacts section below it.
The reviewable output each step produces, such as the search list, the archive list, and the full written summary. This is where a task's final answer lives.
The system saves its state after each completed step. If a task fails or is interrupted, it resumes from the most recent checkpoint β completed steps are never re-executed or re-billed.
The simulation mode used when no API key is configured: the full pipeline runs normally, model calls return simulated content, and no cost is incurred.
5. Frequently asked questions
Q: Where can I view the results of a completed task?
A: Open the Tasks page and click the task to enter its detail page; the results are in the "Artifacts" section below the flow diagram. The same results are also archived in the Library and included in the next Briefing.
Q: What does the flow diagram at the top of the task detail page represent?
A: It shows the real-time status of each execution step of the task; green means completed. The diagram describes the process only β to read the results, scroll down to the Artifacts section.
Q: Why do my tasks keep returning simulated responses without incurring any cost?
A: No API key has been configured, so the system is running in dry-run mode. Paste a valid key on the Settings page to switch to real model calls.
Q: What should I do when a task fails?
A: Open the task's detail page and click "Rerun from checkpoint". The task resumes from the last successful step; completed work is not repeated.
Q: What does a red connection indicator in the top bar mean?
A: The real-time event stream has disconnected, usually due to a backend restart or a network interruption. The browser reconnects automatically; no action is required.
Q: How do I control the system's daily spending?
A: Set a daily budget cap on the Settings page: an alert is issued at 80% of the cap, and tasks are suspended at 100%, so spending can never exceed the cap. The Dashboard shows today's spending in real time, and the Audit page itemizes every call.
For deeper technical details β cost architecture, safety mechanisms and troubleshooting β see the Help page.