### What is FlowithOS
FlowithOS is an Agentic OS that runs inside on your Desktop, no longer bound by just the browser! It uses our self-developed Agent to understand pages, operate interfaces, and complete end to end tasks. It can research, summarize, create content, post on platforms, manage documents, and interact with consumer apps. It runs tasks continuously and can resume after required logins or verifications.
FlowithOS redefines what an Agent can do. Not just a browser, app, or assistant, but a new evolving ecosystem: an Agent that *truly get things done.*
> During the current beta, you are seeing an evolving product. Some functions are still under construction Your feedback will directly shape what ships next.
### When to Use FlowithOS
**FlowithOS vs Neo at a glance**
* **Neo** creates. It is a generator that produces websites, HTML CSS JS webapps, images, videos, text, and assets. It does not operate the sites it creates and it does not control other websites. Use Neo when you want a new artifact that you can open, publish, or hand off.
* **FlowithOS** operates. It is a browser based Agentic OS that navigates pages, clicks buttons, fills forms, collects data, uploads files, and chains actions across sites. Use FlowithOS when you want a task performed on the live web or across multiple apps.
**Give the task to Neo when**
You need a static site or webapp scaffold that runs without Node and installs via a simple zip or GitHub Pages.
* You want content generation, design assets, slide outlines, images, videos, or code that you will test or host yourself.
* You want a reusable template or component library that FlowithOS can later use or reference.
**Give the task to FlowithOS when**
* The job requires logging in, navigating real websites, pressing buttons, uploading or downloading, or filling complex forms.
* You need research with source verification, paging through results, filtering, copying into Docs or Sheets, or exporting CSVs.
* You want cross app workflows such as collect from Site A, transform in Tool B, publish on Platform C, archive to Drive.
* You want continuous or repeated runs such as monitoring a page, refreshing a dashboard, or executing a routine every day.
**Use both together**
* Generate a tool or resource in Neo, then have FlowithOS open it, test it, populate it with data, and use it in a broader workflow.
* Store brand rules or project defaults as Memories and procedural steps as Skills, then run the live task in FlowithOS.
* drag tabs from the sidebar directly onto Neo's canvas to instantly start a conversation
> We're working on creating a more powerful and seamless integration between the Flowith website and FlowithOS. Stay tuned for this update in a future release!
### Quick start checklist
1. Install FlowithOS and open the app. (FlowithOS updates often during the public Beta period. Keep it current for the best results.)
2. Sign in with your Flowith account credentials
3. Open the OS Agent in the bottom right corner.
4. Choose a mode. Fast is best for quick one off tasks. Pro is best for complex pages and detailed layouts, using deeper reasoning, skills, and memories.
5. Enter a prompt that follows this pattern: Location + Task + Criteria + Deliverable.
6. Press Enter.
7. Watch Steps to see what the OS is doing. Open `Files` to review outputs.
> Tip: FlowithOS may need you to log into websites to execute tasks properly. Simply step in when prompted to complete the login, and it'll continue from there.
### The FlowithOS interface at a glance
#### 1. Homepage
* Command Center: type a prompt, then choose Run Task or Search.
* Run Task sends your prompt to the Browser Agent for analysis and execution.
* Search performs a focused web search that gets to key information quickly.
* Preset Tasks: click to load an example into the input box, then Run Task.
* Frequent Visit: shortcuts to your most visited sites.
#### 2. Tabs and navigation
* The sidebar shows an icon for each open page.
* Hover an icon to reveal quick actions: refresh, forward, back, bookmark.
#### 3. Top level sections
* Tasks: view all running and completed tasks.
* Flow: access flows associated with your Flowith account. You can also drag a page icon from the sidebar onto the canvas to start a conversation.
* Skills and Memories: Create Skills and Memories for the Agent to access during task
* Account: sign in to your Flowith account for the full OS experience.
#### 4. OS Agent panel
* Official Preset Prompts To help you jumpstart your experience, we've curated some showcases of expert-crafted prompts. A single click will instantly load a task into the input box. You can then refine the prompt to your specific needs or simply press Enter to launch the task immediately.
* Your Personal Preset Library Beyond our showcases, you are able to build your own library of custom prompts. Click the "Add a new preset" option under the Preset button to save and reuse your most effective commands, tailoring your workflow for maximum efficiency.
* Attachments: upload files to guide the task.
* Text and code: md, txt, json, csv, xml, js, ts, jsx, tsx, py, java, cpp, c, html, css, scss, php, rb, go, rs, swift, kt, sh, bat, yaml, yml, rtf, log, ini, cfg, conf, env, properties
* Documents: pdf, doc, docx, xls, xlsx, ppt, pptx
* Images, media, archives: png, jpg, jpeg, gif, svg, webp, bmp, tiff, tif, ico, mp4, avi, mov, mkv, webm, flv, wmv, m4v, 3gp, mp3, wav, flac, aac, ogg, m4a, wma, zip, rar, 7z, tar, gz, bz2
* Execution Modes:
* Fast mode: prioritizes speed and delivers quick results. Does not use Skills or Memories
* Pro mode: performs deeper visual understanding and layout reasoning for complex interfaces. Also utilizes Skills and Memories
> We recommend using Pro mode as your default for most tasks. Pro mode allows for more complex challenges, Skills, and Memories.
* Execution Details:
* Steps: Gain full transparency into your task's execution. Here, you can watch in real-time as FlowithOS navigates, clicks, extracts information, and inputs data on your behalf.
* Files: This is your dedicated results hub. All files generated during the mission, along with the final deliverables, are consolidated here for instant access.
* Ask Follow-Up or Launch a New task You can continue the conversation by using the input field in the sidebar to ask follow-up questions on the current task. And you could simply return to the main input box at the bottom-right of your screen to begin a completely new mission.
### Your first task
1. Head to the OS Agent in the bottom-right corner and input your command. You can also explore our curated Presets to get a feel for the workflow.
2. Click Presets and select one that matches your goal or type your own prompt.
3. Click Select Mode for the most robust experience, we highly recommend the more powerful Pro Mode.
4. Press and watch as FlowithOS masterfully deconstructs and executes your task.
You will see:
* Steps: a live log of how the OS is navigating, clicking, extracting, and typing.
* Files: files and final deliverables produced during the task.
> You can ask follow up questions in the OS Agent sidebar to refine the result. You can also start a new task any time from the main input in the bottom right.
### Day 1 tips
* Think in outcomes. Tell the OS what to produce and where to put it.
* Add criteria. Quantity, date ranges, quality bars, and formatting rules make results better.
* Start with Fast mode. Switch to Pro mode if the site is dense or highly interactive.
* Give it plenty of context. Say “filter to the last 7 days” or “expand the table to 50 rows”.
* Save good prompts. Anything reusable should live in Your Presets.
### The prompt pattern that works
**Location + Task + Criteria + Deliverable**
* Location: where to operate or where to store results
* Task: the action you want
* Criteria: the rules for quality and scope
* Deliverable: the final output and format
**Example**
> Scan Product Hunt for today’s top 1 product. Review screenshots and descriptions to judge investment potential. Create a concise report in a new Google Doc.
* Location: Product Hunt, Google Docs
* Task: find product, review visual and textual materials, judge potential
* Criteria: top 1 today, analysis must reference screenshots and descriptions
* Deliverable: a Google Doc with a concise report
**Why this works**\
The OS sees where to go, what to do, how to evaluate success, and what to produce. Clear prompts reduce retries and produce higher quality outputs.
### Ready to use prompt ideas
**Research and summarize**
* Find a top-rated restaurant featured on Michelin Guide and Google Maps in New York City, summarize the top reviews, and show me how to book a 2 party table online.
**Competitive scan**
* I need some startup ideas. Open Hacker News, quickly list 5 AI startup ideas you find, and then create high-quality write-ups in Google Docs.
Create a new knowledge base on flowith.io (click \[Knowledge Garden] Button in the left sidebar, then select Add new knowledge Bases, and enter the new knowledge base name. After naming the knowledge base, wait for 10 seconds. Paste the content of the finished Google Doc into the text input box of the named knowledge base, then click Add to complete the import. After the completion, tell user the knowledge base is available for future use.
**Content drafting**
* Select one of the top 10 trending posts on X.com, mimic the style and focus of my past homepage posts, and share it as a quote retweet (click "Repost," then "Quote," add your comment, and finally click "Post" to complete the sharing).
**Search and Plan**
* Find a top-rated restaurant featured on Michelin Guide and Google Maps in New York City, summarize the top reviews, and show me how to book a 2 party table online.
### Skills
Skills are reusable custom instructions written in Markdown that tell the Agent exactly how to navigate a site or complete a task step by step. Think of a Skill as a precise playbook. When your task runs, the OS will reference relevant Skills to plan and execute actions.
**Create or edit a Skill**
1. Open the side menu.
2. Click Create new Skill.
3. Rename the `new-skill.md` file
4. A Markdown composer opens. Write or paste your Skill content.
5. The composer will automatically save
6. To edit later, open the side menu, select the Skill, update the Markdown, then Save.
**Confirm a Skill is being used**\
During task execution, check the pink brain icon to the left of any step. Hover to see the tooltip. The tooltip lists the Skill that influenced that step.
**How to write a high quality Skill**
* Give the Skill a clear name and a one sentence objective.
* Specify exact page locations to operate in.
* Use literal click and type guidance for fragile UI.
* State success criteria and the final deliverable.
* Include guardrails such as date windows, quantity targets, formats, and required confirmations.
* Prefer short, imperative sentences.
**Skill Example**
```md
name: publish_video_on_Youtube description: Publish long videos or Shorts on YouTube website: '' related_skills: []
Publish Video on Youtube
Instructions
Obtain the file ID that needs to be uploaded. This could be the File ID returned from a previous download action, or it could be the ID of a file manually found in the file system based on the upload file description, or it could be a file uploaded by the user as an attachment.
Go to [https://studio.youtube.com/channel/](https://studio.youtube.com/channel/) , click the "Create" button in the upper right corner, and click "Upload Video" in the dropdown menu that appears.
Use the upload_file Action capability to upload the file that needs to be uploaded to the "Select File" element in the modal that pops up. It is strictly forbidden to directly click this element to upload.
Wait 10 seconds until the upload is complete and the modal displays a detailed information form. If it doesn't appear, continue repeating this step.
Fill in the video title and description in the form, then fill in the "Made for Kids" form item according to the specific content of the video and the audience.
If any confirmation bubble appears, click to close it.
In the next few steps, continue clicking "Next" until you reach the "Visibility" step. Select the visibility range as needed, then click "Publish"
```
**Maintenance tips**
* Keep Skills focused on one workflow. Create a new Skill for different sites, or to separate complex task on a single site
* Review Skills that are not triggering. Update instructions or clarify steps as needed.
**Troubleshooting Skills**
* The Skill did not trigger: make sure the Location line matches the active site and that the task prompt mentions the same site.
* Steps are being skipped: try to add words from your skill file to your prompt to help the model
### Memories
\
Memories are persistent facts the Agent should remember across tasks. Use them for context that improves decisions and output quality. Examples include brand voice, formatting preferences, target regions, approved tools, and stable account details that are not secrets.
**Create or edit a Memory**
1. Open the side menu.
2. Click Create new Memory.
3. Rename the `new-memory.md` file
4. A Markdown composer opens. Write the Memory content.
5. The composer will automatically save
6. To edit later, open the side menu, select the Memory, update the Markdown, then Save.
**Confirm a Memory is being used**\
During task execution, check the pink brain icon to the left of any step. Hover to see the tooltip. The tooltip lists Memories that influenced that step.
**What belongs in Memories**
* Brand and tone guidelines with concrete examples.
* Formatting rules such as preferred deliverables, column orders, and naming patterns.
* Stable constraints such as budget ranges, date windows, and safe sources.
* Tool preferences such as where to store outputs and which apps to use first.
* Accepted definitions such as what qualifies as a lead or a win.
**What does not belong in Memories**
* Secrets and credentials.
* Rapidly changing facts such as daily prices or schedules.
* Single use instructions that should live in a Skill or a one time prompt.
**Recommended Memory template**
```md
# Memory name: Marketing defaults for research and reporting
Voice and tone:
- Confident, clear, friendly
- Avoid slang
- Prefer short paragraphs with bullets for key points
Output defaults:
- Primary deliverable: Google Doc summary
- Secondary deliverable: CSV table in Assets when lists exceed 10 rows
- File naming: {yyyy-mm-dd}_{project}_{short-title}
Quality rules:
- Always provide direct links for claims and data
- Prefer sources published within the last 90 days for market data
- When uncertain, ask a single clarifying question before proceeding
Research scope:
- Priority sources: company sites, official docs, primary data
- Acceptable sources: trusted news and analyst outlets with bylines
- Exclude: forums unless explicitly requested
Constraints:
- Budget examples should use USD
- Target regions: United States and Canada
```
**Maintenance tips**
* Keep Memories short and stable. If it grows beyond one page, split it into focused Memories.
* Review Memories monthly and archive what is no longer useful.
**Troubleshooting Memories**
* The output ignores formatting rules: confirm the Memory exists and is active, then restate the key rule in your prompt once.
* Outdated preferences are applied: open the Memory, edit the outdated line, and Save.
* The wrong region or source appears: tighten the Research scope section and add explicit exclusions.
### Pro tips
* Monitor the output of your task, then build skills and memories accordingly to avoid future mistakes or to replicate success.
* Use precise Locations. If an output must land in Google Docs or Sheets, say so.
* Set guardrails in Criteria. Define minimum counts, date windows, formats, and sources.
* Chain deliverables. Ask for a summary, then a slide outline, then a final deck draft.
* Attach context. Upload brand voice docs, target lists, CSVs, and style guides for higher fidelity.
> while the task is running you can send follow-up prompts to send more instructions and the Agent will pause immediately to read your instructions
### Troubleshooting
* The OS stops to ask for a login\
\- Sign in directly on the site. The OS cont qinues once authentication completes.
* The site has infinite scroll or complex UI\
\- Switch to Pro mode. Add explicit Criteria such as number of items to load.
* The result is close but not perfect\
\- review your prompt and the models output, then decide if the skill, memory or prompt needs updating
### Thank You!
FlowithOS is a collaborative effort with our Pioneer community. We have tested the OS across posting, ordering food, trading, chatting, and gaming. The most valuable examples come from real work and real creativity. Share what you discover, and we will keep building with you.
Enjoy the flow.
---
# Agent Instructions: Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.
Perform an HTTP GET request on the current page URL with the `ask` query parameter:
```
GET https://doc.flowith.io/flowithos/copy-of-flowithos.md?ask=
```
The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.
Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.