Frequently asked questions

The easiest way is to select your files in a desktop app like Riflet, which bundles them into a single structured text file. You then paste or upload that file into ChatGPT. This works because ChatGPT accepts plain text files up to 512MB. The advantage over uploading files one by one is that you get all your project context — notes, docs, PDFs, code — in one upload, and you can save that bundle to reuse in future sessions without rebuilding it each time.

Claude accepts up to 30MB per file. If you want to feed multiple files — say a folder of research notes, a brief, and several reference documents — the practical approach is to merge them into one structured text file first. Riflet does this locally on your computer: you select files and folders, and it outputs a single file you can upload to Claude. Claude Projects also lets you upload files that persist across sessions, but this only works in Claude — not in ChatGPT or Gemini.

Neither ChatGPT Custom GPTs nor Claude Projects are portable between platforms. If you want the same project context available in multiple AI tools, the cleanest solution is to build a reusable bundle — a single structured text file containing all your relevant files — and paste or upload it at the start of each session in whichever tool you're using. Riflet saves these bundles locally so you can reload them in seconds rather than reassembling the files each time.

Because AI models have no memory between sessions. Each new conversation starts completely blank — the AI doesn't know what you told it yesterday. The only context it has is what you give it in that specific session. This is why power users build context bundles: a saved collection of files (project notes, background documents, style guides) that they paste at the start of every session to give the AI immediate, consistent knowledge about their work.

Save your project context as a reusable bundle. This typically includes: a brief description of the project, any relevant documents or notes, key constraints or decisions already made, and the tone or style you want the AI to use. Tools like Riflet let you select these files once, bundle them, and reload the bundle in any future session — in any AI chat tool. You paste once at the start, and the AI immediately has full context without you having to re-explain anything.

ChatGPT allows up to 10 files per message on Plus plans, with a rolling limit of around 80 files per 3-hour window on the free tier. A common workaround is to combine all your files into a single text file before uploading — that way you're uploading one file that contains the content of many. Tools like Riflet do this automatically: select a folder, and it merges all the files into one AI-ready output. ChatGPT processes the combined file as a single upload, giving you access to all the content at once.

Select the specific folders or notes you want the AI to know about — your active projects, relevant MOCs, character or reference notes — and bundle them into a single text file. Riflet can read your Obsidian vault as a local folder, letting you pick exactly which notes to include. The result is a Markdown-formatted bundle you paste into ChatGPT, Claude, or any other AI. Unlike Obsidian plugins that talk to one specific AI, this approach is model-agnostic: the same bundle works everywhere.

In Notion, open Settings, then Developer tools, then Connections. Create a new connection and copy its Internal Integration Secret, which starts with ntn_. Open the Content Access tab in that connection and select the pages and databases you want to share, otherwise the connection can read nothing. Then in Riflet, add a Notion source and paste the secret. Riflet encrypts the token with a key held by your operating system and lists your workspace as a tree you can pick pages from.

Export the relevant Notion pages as Markdown (Notion → Export → Markdown & CSV). Then use Riflet to select those exported files and bundle them. The resulting file can be pasted into any AI chat. This gives you a snapshot of your Notion workspace that the AI can reason over, with no setup at all. Riflet can also connect directly to a Notion workspace using a Notion connection token, if you would rather pull pages in without exporting them first.

A context window is the maximum amount of text an AI model can read and consider at one time — essentially its working memory for a single conversation. It is measured in tokens, where roughly 750 words equals 1,000 tokens, though the newest models tokenize more finely than that. Claude Opus 5 has a 1,000,000-token context window (around 555,000 words). ChatGPT has a 1,050,000-token window on GPT-6 Astra and on the GPT-5.6 models. When your conversation or uploaded files exceed this limit, the AI starts losing track of earlier content.

It depends on the size of your files. As a rough guide: 1,000 tokens is about 750 words, or 3–4 pages of text. Claude Opus 5's 1M context window can hold roughly 555,000 words, the equivalent of several full-length books. ChatGPT's 1.05M window holds a comparable amount. A typical project bundle of 10–20 well-curated notes or documents (without images or binary files) usually fits comfortably within either window. Riflet shows you a token estimate before you export, so you can check whether your bundle fits before pasting it in.

Yes. AI models like ChatGPT and Claude do not retain memory between separate conversations. When you close a chat and start a new one, the model has no knowledge of what you discussed before. The exception is persistent features like Claude Projects or ChatGPT's memory feature, which store some information between sessions — but these are platform-specific and limited. Reusable context bundles are the model-agnostic solution: you bring the context to the AI rather than expecting the AI to remember it.

Context engineering is the practice of deliberately deciding what information goes into an AI's context window, in what format, and in what order. The term was popularized by AI researcher Andrej Karpathy in 2025 to describe a more intentional approach than just writing prompts. A good context engineering practice might include: a project brief, relevant documents, past decisions, tone guidelines, and key constraints — all assembled before the conversation starts. Riflet is a context engineering tool: it helps you build and reuse context bundles without writing code or using a terminal.

Riflet is a native desktop app for macOS, Windows, and Linux that lets you build reusable context bundles from your local files and folders. You select the files you want — notes, documents, PDFs, code, web pages — and Riflet combines them into a single structured text file. You then paste or upload that file into any AI chat: ChatGPT, Claude, Gemini, or any other tool. The bundle is saved locally so you can reload it in seconds at the start of any future session.

Claude Projects and Custom GPTs only work within their own platform — you can't use a Claude Project in ChatGPT, and vice versa. Riflet is model-agnostic: the same bundle works in every AI tool. Riflet also runs entirely on your machine with no cloud processing, no account required, and no files ever leaving your computer. It also handles files that those platforms don't, including local folders, PDFs, and websites together in one bundle.

Yes. Riflet outputs a plain text or Markdown file that works with any AI tool that accepts text input or file uploads. This includes ChatGPT, Claude, Gemini, Perplexity, Mistral, and local models. Because it uses your clipboard or a standard file upload, there are no integrations to configure — it works the same way everywhere.

No. Riflet processes everything locally on your computer. Your files are never uploaded to any server. There is no account, no login, and no cloud processing. This is particularly important for users working with sensitive documents — client files, research notes, proprietary code, or personal information. The only time content leaves your machine is when you choose to paste or upload it into an AI chat yourself.

Riflet supports plain text files (TXT, MD, CSV, JSON, YAML), code files across all major languages, PDFs, and web pages (via URL or sitemap). It also integrates with GitHub repositories, including private repos on the Pro plan.

Yes. Your Obsidian vault is a folder of Markdown files on your computer. Open that folder in Riflet, select the notes and subfolders you want to include, and export them as a bundle. Because Riflet reads standard Markdown, all your note content, headings, and links are preserved in the output. The bundle works in any AI chat, not just tools with specific Obsidian integrations.

The Free plan lets you work with a single folder, scrape single web pages, and export to TXT, MD or JSON. It covers basic use cases and never expires. The Pro plan adds multiple sources in one bundle (combine a local folder, a website, and a GitHub repo in one export), sitemap import, private GitHub repos, saved workflows so you can reload your bundles in one click, and content minification to reduce token usage. Pro is a one-time purchase — no subscription.

No. Riflet uses a one-time purchase model. You pay once for a Pro license and use it forever. There is no monthly or annual fee. An optional license update ($12/year) lets you receive new features and improvements after the first year, but the core app continues to work without it. This model reflects the fact that Riflet runs entirely on your machine — there are no servers to maintain or API costs to cover.

Repomix and similar tools (code2prompt, files-to-prompt) are command-line tools designed for developers who are comfortable with a terminal. They work only with code repositories and require installing Node.js or Python. Riflet is a GUI desktop app that anyone can use without touching a terminal. It also handles non-code files — PDFs, notes, documents, web pages — making it useful for researchers, writers, and knowledge workers, not just developers.

Yes. If you run a local model (Ollama, LM Studio, or similar), Riflet's output is just a text file you paste into the prompt. It works the same way as with any cloud AI tool — model-agnostic by design.

Only for features that need it: web page scraping and sitemap import fetch content from the web, and the GitHub integration connects to the GitHub API. Everything else — reading local files, building bundles, exporting — works completely offline. There is no connection to Riflet's servers at any point.

Build a "story bible" in Riflet — character sheets, worldbuilding notes, plot outlines, style guides, and draft chapters. Upload the exported file to your AI at the start of each session. The AI gets your full manuscript context, so it stops forgetting character details or changing tone mid-chapter.

Add a folder containing your PDFs as a source in Riflet. It extracts readable text from PDF, DOCX, and 100+ other file types. Select the papers you need, combine them with your annotations or notes, and export one file you can upload to Claude, ChatGPT, or any AI for synthesis and literature review.

A useful context bundle typically has four parts: (1) a project brief — what the project is, its current status, and key constraints; (2) reference documents — notes, research, past work, style guides; (3) people or entity information — who is involved and what matters about them; (4) instructions for the AI — what kind of help you want and what to avoid. Keep bundles under 50,000 words so they fit comfortably within most context windows and the AI doesn't lose focus.

Update it at the start of a new phase or when the core facts change — not after every conversation. A good bundle represents the stable knowledge about a project, not a real-time log. Many users have one bundle per active project, updated every week or two. The goal is that the AI immediately understands your project the moment you paste the bundle, without you needing to add any additional explanation.

Yes. Riflet lets you save multiple workflows, one per project. Each workflow remembers which files and sources you selected, so reloading a project bundle for a new AI session takes a single click. There is no limit to the number of saved workflows on the Pro plan.

Both work, but Markdown is generally better. AI models like Claude and ChatGPT handle Markdown well — they understand headers, bullet points, and code blocks, which helps them navigate the context quickly. Riflet exports in Markdown by default, with clear section headers separating each source file. This makes it easier for the AI to locate specific parts of your context when answering questions.