My phone’s Downloads folder has 379 files in it. I know that only because I asked.Finding anything 2026-10-27 20:25:7 Author: hackernoon.com(查看原文) 阅读量:23 收藏
My phone’s Downloads folder has 379 files in it. I know that only because I asked.
Finding anything in there used to mean the same routine: open a file manager, sort, scroll, squint. Now I just ask.
The app doing this is Gotcha, the open-source Android copilot we build at Samosa AI. It is not a file manager and does not try to be one. It is a copilot with 110+ tools: it makes calls, sends messages, sets alarms, edits media and PDFs, runs terminal commands, and works inside other apps.
Files are just one of the things it handles, but that part does the job people usually go looking for in an AI file manager: you ask about your files in plain English and get an answer, not a folder to dig through.
So I put that to the test. I asked it four questions on my Nothing Phone 3a, checked every answer, and then looked at what you would use instead.
If you are new to Gotcha, start with Gotcha: The Android Copilot That Actually Works, which explains how it turns plain English into real actions on your phone.
The idea in one sentence.
A file manager is built around a question you already know how to ask: where is the file called X? Most of the questions I actually have are different. What is taking up space? What did I save last month? Which of these documents mentions Y? What is in this zip? A file manager can help with all of them, but it hands you a list and the thinking is yours.
Gotcha takes the question in plain language and decides which tools to run. For files, that means tools for listing, reading, globbing, and searching file contents with regular expressions, alongside everything else in the tool catalog. It runs in two modes. Monitor is read-only: the agent gets 30 tools that look and none that write, so it cannot change anything even if it wanted to. Operator has the full set. Everything below ran in Monitor unless I say otherwise.
You also don’t have to ask in English. Gotcha’s interface ships in nine languages, and you can ask your question in any language the model you picked understands; the file tools work the same either way.
Question 1: “What are the 10 biggest files in my Downloads folder?”
Twenty-eight seconds. One tool call: a recursive listing of Downloads (379 entries), after which the model ranked them by size and gave me a top ten with sizes, including two videos sitting inside a subfolder I had forgotten about.
I checked it: all ten files, in the right order, with the right sizes.
A file manager can do a version of this too; Files by Google has a “Delete large files” card under Clean. The difference is shape: I asked a question and got an answer that started with “all over 75 MB, topped by a 242 MB APK,” which is the sentence I actually wanted.
Question 2: “Find every PDF I downloaded last month.”
Gotcha came back with 13 PDFs, each with its date and size. I checked: 13 out of 13, correct dates.
Getting there took Operator mode. I asked again with “don’t change anything” added. A direct shell find hit a permission error, so Gotcha handed the job to a sub-agent, which ran the same search through Termux and returned the list in about a minute and three-quarters.
Question 3: “Which text files in my Downloads mention Duolingo? Show me the matching lines.”
Forty seconds. Two files, seven matching lines, each with its line number, grouped by file. Both files and all seven lines were right.
Gotcha listing seven matching lines across two files.
One line came back as mojibake. I went and looked; the garbage is in the file itself, a box-drawing chart that was double-encoded before it ever reached my phone.
This is where a classic file manager stops being useful. They are built to search file names. Searching inside files, and showing the matching lines with context, is a different job.
Search finds the file. Gotcha finds the line in that file.
Question 4: “What’s inside this zip? Open one of the images and describe it.”
Gotcha listed all seven entries of a 21 MB archive, then read one of the PNGs straight out of the zip, without extracting it, and described it in detail. I compared it with the image itself, and the description was accurate.
Any good file explorer can open a zip.
What I could not do before was ask a question about a picture that was still inside one.
Question 5: “Take Award-session.mp4 from that list, pull out the audio, keep only the first two minutes, and transcribe it.”
The biggest-files list had a 177 MB video in it, so I asked a follow-up. This one needs Operator mode, because it creates files.
Fifty-seven seconds, four tool calls. Gotcha listed the folder to find the video, extracted its audio track into a new file (2:58, 4.1 MB), trimmed the first two minutes into a second file, and sent that clip to the speech-to-text provider I had configured. The original video was not touched.
The trimmed clip is exactly two minutes long and sounds identical to the start of the video, because the extract and the trim copy the audio stream instead of re-encoding it, so there is no quality loss.
The transcript reads cleanly and gets the substance right: 94 submissions, 28 accepted papers, the award, the paper’s title, and all seven authors in order. Proper nouns are where it slips. “Hans Meuer” became “Hans Moyer,” the tool sys-sage became “CISSAGE,” and most of the author surnames came out spelled by ear. That is the speech-to-text model, not Gotcha, and it is worth knowing before you paste a transcript anywhere that matters.
Two minutes of audio did leave the phone for this step, to the speech-to-text endpoint, and the tool’s own output says so. The extraction and the trim happened on the device.
This is the part a file manager is not built for. The question and the follow-up work happen in the same conversation, using the media and PDF tools that arrived in v1.2.0.
While you wait.
Answers took 28 to 40 seconds. A file manager’s search is instant, so if you know the filename, use the search box. For everything else, you do not have to watch: since v1.3.0, Gotcha keeps working in the background when you switch apps and sends a notification when it is done or needs you.
If you haven’t read about the latest Gotcha version yet, have a look at Gotcha v1.3.0: Your Android Copilot Just Got a Major Update, which also covers on-demand permissions, chat backups, and the new interface languages.
Coming soon.
File support is growing quickly. Coming soon: file dates in Monitor mode, reading PDFs straight from storage (attaching a PDF in the chat already works), cleaner Word, Excel, and PowerPoint reading, dedicated move, rename, and delete tools, RAR and 7z archives, and an on-device index for fast photo search. Current limits are listed in the FAQ.
One thing to know today: in Operator mode, moving and deleting files happens through shell commands that run without a confirmation prompt. Use Monitor mode for exploring, and a file manager with a trash bin for bulk deletes.
How it compares.
When I started this post, I assumed nothing else on Android does this. That turned out to be wrong, and it is worth being precise about who does what.
Classic file explorers (Files by Google, Solid Explorer, Material Files) are excellent at what they were designed for. Solid Explorer opens RAR, 7z, and TAR archives. Material Files is open source under the GPL. Files by Google finds large files across the whole device, and on supported setups lets you ask Gemini about a PDF you are viewing (under 25 MB, not password-protected). They are instant, and their move, copy, and delete flows have years of polish. They do not answer open questions about your files. The Gemini app answers questions about files you upload into a chat, but does not browse your phone’s storage.
AI file managers are the real comparison, and there are several on the Play Store now. These descriptions come from their store listings as of October 2026:
- Smart File Manager does natural-language search with dates and types built in, for example “PDFs from January” or “Large videos over 500 MB,” with filtering on the device.
- AI File Manager builds an opt-in on-device index of your documents and photos, reads inside PDF, Word, Excel, and PowerPoint files, and supports search by meaning, which makes it strong at “find the document about X” and at photo search.
- PureFlow is the closest to Gotcha in spirit: a chat agent that works with your own OpenAI-compatible model API, finds, moves, and measures files, asks before deleting, and also handles a few phone tasks like alarms and calendar events.
So where does Gotcha actually fit?
Gotcha is not a replacement for your file manager, and it is not meant to be. It is a phone agent, and the filesystem is one of the surfaces it works on, next to calls, messages, the calendar, media and PDF editing, the terminal, and other apps.
That is what makes it useful with files: a question can turn into action in the same conversation. Find the biggest recordings, trim one, transcribe it, share the transcript. Pull the matching lines out of three research exports and draft a summary from them. List what is in a folder before you hand it to a PDF merge.
If you missed it, we wrote about the media side of this in
Other apps added AI to a file manager. Gotcha is a phone copilot that happens to be good with files.
Three properties make that trustworthy enough to try.
- Monitor mode is a real boundary: the read-only agent is never given a tool that writes.
- The whole app is open source under the AGPLv3, so every claim in this post can be checked against the code.
- And you choose the model. File operations always run on the device; what Gotcha reads to answer you goes to the model you picked. Pick a local model, and nothing leaves the phone.
Try it.
Gotcha runs on Android 11 and above, free and open source. The interface comes in nine languages (English, Hindi, Spanish, French, German, Italian, Portuguese (Brazil), Japanese, and Simplified Chinese), and you can ask questions in any language your model understands.
- Source code: https://github.com/samosa-ai-com/Gotcha
- Landing page: https://samosa-ai.com/gotcha
- Download the APK: https://github.com/samosa-ai-com/Gotcha/releases/latest
- Every tool, by capability: https://samosa-ai.com/gotcha/docs/tool-catalog
Start in Monitor mode, grant “All files access,” and ask it the question about your phone that you have been meaning to answer.
More from the Gotcha blog
如有侵权请联系:admin#unsafe.sh