My workflows
How I actually use SanFrame
When I started this project, my original need was pretty simple: take a screenshot, get its file path right away, and paste it straight into my terminal so AI can read it while we chat. Since then, I’ve also been using it to save good web content, dictate across machines, pass specs and screenshots around when developing across systems, and let terminal AI save organized modification points back into SanFrame via MCP, group them, and pin them on top. Here are the five workflows I use most.
sansan · My five everyday workflows
01Capture → Terminal AI
Take a screenshot, get its path right away, and paste it into the terminal
This was my original reason for building SanFrame: I like developing with terminal tools such as Codex CLI and Claude Code, and I wanted the images I drop in to be read directly by AI in the terminal.
Now, right after I take a screenshot, the image stays in Image tools and its path goes straight to my clipboard—or I annotate it first if I need to point something out. I just paste the path into the terminal and start chatting with AI.
Capture → Get local path → Paste into terminal
Workflow diagram
Paste the path straight into a terminal AI tool that can read local images, then type your question.
Sample content illustrates the author’s actions, not a live transfer or an AI response. Paths are examples only.
This computer
Capture the screen
SanFrame · Image tools
Original path copied by default
C:/Demo/capture.pngExample local pathThis computer · AI terminal
Conversation input
C:/Demo/capture.pngTake a look at this image—how should we adjust the button spacing here?
Ordinary captures copy the original path by default. The terminal tool must be able to read local images. Images shared with an online AI service are handled under that service’s policies.
02Web → OCR & Pin
See something good on a website? Just take a screenshot
When I see good content on a website, I just take a screenshot. Afterwards it’s automatically saved in the Image tools list, and with auto OCR enabled, the text inside gets recognized too.
If it’s important, I pin it so it sits right at the top next time I open the list. Even if I only remember a few words later on, I can search the recognized text to pull up the original screenshot.
Saved to the list automatically, pinned when it matters
Just capture it
Web content is saved straight into Image tools
Auto OCR when enabled
Recognizes the text so you can search it later
Pin what’s important
Stays on top when you reopen the list
03Mac Voice → Win Terminal
Speak into one computer, and it lands right on the other clipboard
I usually have two computers open—say one Windows PC and one Mac, though more machines can join and you can pick one or several recipients. On my Mac I use the open-source project Handy for speech-to-text; it already does enough that I don’t need to reinvent the wheel, so I don’t install it again on Windows.
There was a real pain point here on Windows: I like developing with terminal tools (such as Codex CLI) rather than the ChatGPT desktop app, and at the time voice input couldn’t go straight into the Windows terminal (I don’t know whether that’s been solved now, but that was the case back then). So I just speak directly into the Device transfer input on Mac, and when the text reaches the other computer it goes straight to the clipboard, ready to paste into the terminal.
Dictation across devices: Mac voice → Windows clipboard
Workflow diagram
On Windows, paste that same text straight into the terminal AI conversation input.
Sample content illustrates the author’s actions, not a live transfer or an AI response. Paths are examples only.
Mac · Handy
SanFrame · Device transfer input
Let’s tighten the spacing here so I can test it on Windows right away.
Windows · Device transfer
Let’s tighten the spacing here so I can test it on Windows right away.
Copy received text · On
Let’s tighten the spacing here so I can test it on Windows right away.Windows · AI terminal
Conversation input
Let’s tighten the spacing here so I can test it on Windows right away.Run SanFrame on the same LAN and connect the computers. Enable Send live on the sender and Copy received text beforehand on the recipient (off by default).
04Cross-System Sync
When building across systems, fix and test right where the bug shows up
When developing software that runs on both Windows and Mac, convenient transfers between devices become really useful. If a problem shows up while testing on Windows, I fix and test it on Windows; if an issue appears on Mac, I develop, patch and test it on Mac—multiple Windows and Mac computers can all work together.
Along the way I often need to pass requirements and screenshots across machines. For example, when I take a screenshot on Mac, it syncs over to the other device and puts that device’s saved local path onto the clipboard, so I can paste it straight into the terminal there.
Across systems: Mac capture synced → Windows path in terminal
Workflow diagram
Paste that Windows local path into the terminal AI tool on Windows and keep investigating.
Sample content illustrates the author’s actions, not a live transfer or an AI response. Paths are examples only.
Mac · New capture
Within the selected sync scope
Windows · Image tools
Receiving: image paths · On
C:/Demo/received/mac-capture.pngExample receiving-side pathWindows · AI terminal
Conversation input
C:/Demo/received/mac-capture.pngHere’s the bug screenshot synced from Mac—help me check what’s causing it.
Connect the computers on the same LAN, enable image sync and choose a scope. Separately enable image-path copying under Receiving on the recipient (off by default).
05Terminal AI → MCP & Pin
Finish planning in the terminal, and let AI save and pin the changes in SanFrame
Besides pasting screenshot paths into the terminal for AI to read, I often use it the other way around: after tracking down a bug or figuring out what to change next with Codex CLI or Claude Code, the terminal usually holds several conclusions, things to fix, and file paths we just touched. If they stay buried in a long terminal session, they take a while to dig up once I close the window or come back a couple of days later.
With SanFrame’s MCP connected, I just tell AI in the terminal: “Save these modification points we just discussed—along with the touched file paths and the current terminal session ID—as separate entries in Pasted content, group them into a ‘To-fix list’ group, tag them ‘To-do,’ and pin them.” (If I also want the files listed separately, I can ask it to add them to File references with the same tag.) When I switch back to the desktop or open SanFrame a couple of days later, the pinned items sit right at the top, with the modification points, file paths and session ID ready to resume.
Ask in terminal → Saved to Pasted content via MCP → Grouped, tagged & pinned
Workflow diagram
Pinned at the top of the list so you can grab the modification points and session ID right away when you return.
Sample content illustrates the author’s actions, not a live transfer or an AI response. Paths are examples only.
This computer · AI terminal
Connected to SanFrame MCP
Session: 8f3a2cSave our modification points and session ID into Pasted content, group as ‘To-fix list’ and pin
SanFrame · Pasted content
Same-workspace group “To-fix list” · #To-do
Session: 8f3a2c · src/workspace.rsExample session ID & noteSanFrame · Top of list
Pinned group & context
Session: 8f3a2c · src/workspace.rsOpen the list anytime to find the pinned “To-fix list” group and #To-do tag with your notes and session ID.
Enable External AI access from AI access in the top toolbar, grant the target workspaces along with record creation, groups, tags and pins, and keep SanFrame running. AI can group at least two records within the same workspace (such as multiple pasted notes or multiple file references), not across images, text and file references. Every AI change is logged in Operation history for review or undo.
A side note: why I prefer the terminal over desktop clients
Earlier I mentioned preferring terminal tools over desktop clients (for example, using Codex CLI rather than the ChatGPT or Codex desktop app). Besides personal habit, I had seen community reports about heavy network and memory usage in desktop apps, whereas working in the terminal uses fewer resources:
Community reports checked on November 14, 2026
- A Windows user reported about 31 GB downloaded over roughly 58 hours. Network report
- Another report described desktop-process memory reaching about 8.8 GB (that issue was later closed). Memory report
Of course, these are individual reports from specific versions and setups, not a claim about current behavior or a terminal-versus-desktop benchmark.