How they fit together
A Gem holds your context and your instructions. Riflet builds the file you put in it. Setting up the Gem takes minutes. Gathering the source material and cutting it down to what matters is the real work, and that is Riflet's half.
These are not competing tools. Gemini Gems is built by Google and is where your context lives. Riflet is a local-first desktop app that assembles that context from mixed sources and exports it as one portable file. You use both.
Who does what
| Task | Riflet | Gemini Gems |
|---|---|---|
| Collecting sources from repos, sites, vaults, and folders | Connects to each source type and pulls the files together in one workspace. | Not part of what it does. It receives files you already have. |
| Choosing what the AI should and should not see | Visual file tree, preview, ignore patterns, and saved filter presets. | You decide when attaching knowledge, one file at a time. |
| Reading PDFs, DOCX, XLSX, and PPTX into clean text | Parses them locally into Markdown and fenced code blocks. | Handles common documents directly once attached. |
| Knowing the size of the context before you use it | Live token count against the selected model window, before export. | You find out once the knowledge is attached. |
| Saving instructions and behaviour for repeat use | Not its job. Riflet saves the source setup, not the prompt. | This is exactly what a Gem is for. |
| Using the same context in Claude or ChatGPT too | The exported file is plain text, so it uploads anywhere. | A Gem and its knowledge stay inside Gemini. |
What Gemini Gems does well
- A Gem saves your instructions once, so you stop repeating the same setup at the start of every conversation.
- Knowledge files attached to the Gem are available in every chat with it.
- It sits directly inside Gemini, with no extra step once it is configured.
- It is the right place for a repeatable assistant if Gemini is your main tool.
What it leaves to you
- A Gem does not collect sources. A repository, a documentation sitemap, an Obsidian vault, or a Notion export has to be turned into files first.
- Knowledge files are attached and updated by hand as your sources change.
- There is no preflight view of what the attached knowledge costs in tokens.
- A Gem lives in Gemini. The same context has to be prepared separately for Claude or ChatGPT.
Building the file in Riflet
Add your sources in Riflet
Local folders, a public or private GitHub repository, a web page or full sitemap, an Obsidian vault, or a Notion export can all live in the same workspace.
Select exactly what the Gem should know
Use the file tree, preview, and ignore patterns to keep the knowledge focused. A Gem with tight knowledge answers better than one holding everything you own.
Check the token count before you export
The live token count shows the bundle size against your model's context window while there is still time to trim.
Export one file and attach it as Gem knowledge
Export a single .txt or .md, then attach it when you create or edit the Gem. Write the Gem instructions around the file's structure.
Build your Gem knowledge file with Riflet
Combine files, folders, repos, websites, PDFs, Office docs, and notes into one context file, then upload it where you already work.
Download Riflet freeFrequently asked questions
Do I still need a Gem if I use Riflet?
Yes. A Gem stores the instructions and keeps the knowledge attached for every conversation. Riflet only builds the knowledge file. It has no instructions, no chat, and no model behind it.
Does Riflet update Gem knowledge automatically?
No. Riflet creates the export and you attach it to the Gem yourself. When sources change you re-export and replace the file. Saved workspaces make the re-export quick, but the attachment stays manual.
What should go in the Gem instructions versus the knowledge file?
Instructions describe how you want Gemini to behave, its role, tone, and output format. The knowledge file carries the facts, the source material the answers should be grounded in. Riflet handles the second one.
Can I use the same file in Claude or ChatGPT?
Yes. The export is plain text, so the same bundle works in Claude Projects, ChatGPT Projects, Perplexity, and local models without being rebuilt for each platform.