Repo Prompt 2.1.33 — Mac version

You've got an AI agent wired into your codebase. It's half-useful, half-unhinged, because the context window is a mess of old errors, duplicated files and a readme from three commits ago.

1. Build a prompt that knows the job. Use the prompt library to save instructions you actually need again — lint rules, file layouts, agent behaviour notes — and drag them in instead of typing from scratch.
2. See what you're spending. The real-time token counter shows how much of your context budget each file and instruction eats. Trim the fat before the agent starts guessing.
3. Let the tools share a brain. With MCP integration, multiple tools can pull the same file analysis and context. No more pasting the same class definition into three different windows.

Because it runs natively on macOS, the app doesn't add a browser tab to your already groaning window bar. It sits alongside Xcode or your terminal and does its one job: give coding agents a cleaner brief so they waste fewer turns.

Picking files to shape the context you send to an AI

Repo Prompt isn't trying to replace your editor or your terminal. It lives between them and the AI models, acting as a context middleman. You point it at a repository, tick the files the agent needs to understand, and it assembles a structured prompt before a single token hits the model.

On the macOS side, it plays nicely with the clipboard and standard file dialogs. Drag a folder from Finder onto the file tree and it picks up the structure. The app also respects your system appearance setting — if you run everything in dark mode, it blends in without a fuss. Because it's a native app, ⌘C and ⌘V work as you'd expect when moving snippets between Repo Prompt and your editor, which isn't always true of web-based context tools.

With MCP wired up, it becomes a shared context layer. Your code-review tool, your test runner and your chat agent can all pull the same curated file set. You stop being the clipboard courier for class definitions, which alone saves ten minutes of faff per session.

Bringing your own keys for Ollama or cloud models

Does it need its own API keys? You bring your own keys for whatever models you use — local or cloud. The app doesn't bundle any model access. If you point it at Ollama running on your Mac, it'll talk to that. If you use a cloud provider, you paste in those credentials once.

Will it slow down a large repo? The file-tree browsing and search are built on macOS-native file APIs, so a project with a few thousand files scrolls fine. Where things get heavy is the AI work itself, and that depends on the model you pick, not the app. If you're checking out a repo prompt crack mac build, you'll find the performance is straightforward — it's a native app, not Electron wrapped in Python wrapped in hope.

Can it handle more than one repo at once? Yes. You can open multiple repositories in separate windows or switch between them in the same session. Handy if you're working on a backend and a frontend that share an API contract but live in different folders.

Stopping the copy-paste loop between tools

Developers who bounce between several AI coding tools get the most mileage. If your workflow involves pasting the same file into a chat window, then into a code-review tool, then into a second model for a second opinion, Repo Prompt cuts out the repetition. The prompt library means you stop rewriting "here's the file structure, don't touch the database layer" every morning.

It also suits anyone who's burned through their token budget debugging nonsense replies. Watching the token counter tick up as you select files teaches you which parts of your codebase are context hogs. A few sessions of that and you start writing terser, cleaner prompts by reflex.

If you mostly work in a single IDE with a built-in AI pane and never touch other tools, you might not feel the need. But for the multi-tool, multi-model crowd — the ones with four terminal tabs and a clipboard manager working overtime — it earns its keep quickly.

• A transition notice now links to the claim portal and next-chapter update.
• Long agent-mode crawling refreshes faster while a run is active.
• Transcription jumping was cut back by stopping auto expansion of tool-card animations.

• macOS 14 or newer
• Runs on Intel and Apple Silicon Macs
• Interface language: English