Documentation

You describe what you need. Kosmo turns it into the perfect prompt for your AI tool.

Kosmo is an AI prompt compiler. Instead of spending 20 minutes crafting a perfect prompt, you type what you want in plain English — "build me a dashboard with auth and Stripe billing" — and Kosmo compiles it into a structured, tool-aware prompt specification that's grounded in the latest official documentation for whichever AI tool you're targeting.

Think of it as a translator between your brain and your AI. You bring the intent; Kosmo handles the prompt engineering, fact-checking, guardrail injection, and formatting. The output is ready to paste into ChatGPT, Claude, Cursor, or any supported tool for instant, accurate results.

Kosmo is not a chatbot. It doesn't answer your questions — it produces deployable specifications that make your existing AI tools dramatically more effective.

Getting started with Kosmo takes about 30 seconds. Here's the walkthrough:

1. Sign up — Join via the waitlist or sign in with your Google account. No credit card required. You get 8 free compiles immediately.
2. Write your first request — Type what you need in the input console. Be as detailed or as vague as you want. Example: "Build a Next.js API route that handles Stripe webhooks with proper signature verification."
3. Choose your target tool — Select which AI tool you'll paste the output into (Claude, ChatGPT, Cursor, etc.). Kosmo adapts the output format accordingly.
4. Hit Compile — Kosmo processes your request in ~15 seconds. It fetches live docs, applies guardrails, and returns a polished prompt specification.
5. Copy & paste — Take the compiled output and paste it directly into your AI tool. Watch the difference in quality.

A compile does more than pass your words straight to an AI model. Knowing roughly what happens in between helps you write better inputs and read the output with more confidence:

• Kosmo works out what you actually want to build or accomplish, including things you implied but didn't say outright.
• It checks current, official documentation for whatever tools, libraries, or frameworks are relevant, so the result isn't based on stale training data.
• It applies the professional practices an expert would use by default, security, error handling, sensible structure, even if you didn't ask for them directly.
• For anything complex, it structures the output so your target AI reasons through the problem instead of guessing at an answer. See Loop Engineering for more on this.
• It formats the result the way your specific target tool actually expects it and hands you a copy-ready output. Each compile costs exactly 1 credit, whatever the complexity.

Not every question belongs in your main thread. Maybe you want to ask something basic without feeling judged for it ("why does this use Next.js instead of plain HTML?"). Maybe you just want to think out loud, vent about a bug that's been driving you crazy, or go on a tangent that has nothing to do with what you're actually building. None of that should cost you the focus of your real work. That's what Vent Mode is for.

Vent Mode is a mirror of your main chat. It sees the same context, so you never have to re-explain what you're working on. But it's a separate space, so nothing you say there touches or derails your main thread. It's a real conversation, not a one-way inbox: you can talk back and forth with Kosmo in Vent Mode the same way you would in your main chat.

What Vent Mode does notdo is compile a prompt. That's not what it's for. It exists so you have somewhere to clear your head, ask the "obvious" question, or process a frustration, without any of it leaking into or cluttering the thread where you're actually building something.

Nothing you say in Vent Mode gets saved anywhere automatically. If something you said turns out to be worth keeping, a real idea worth coming back to, you choose to save it to your Idea Library yourself. Venting and saving are two separate, deliberate actions.

Vent Mode never touches your compile credits. It's free, every time, because it was never meant to produce a compile in the first place.

When you're building something, ideas do not arrive one at a time on schedule. They show up constantly, usually while you're in the middle of something else entirely. Chasing every one of them the moment it occurs is a good way to never finish anything. The Idea Libraryis where you park an idea that's genuinely good, but whose timing isn't right yet, so you can stay focused on what you're actually doing and come back to it later.

Nothing lands in your Idea Library automatically, not from your main chat and not from Vent Mode. You have to manually choose to save an idea, from wherever it came up. That's deliberate: the library is for ideas you decided are worth keeping, not a transcript of everything you ever typed.

Every idea has a status:

• Active, a fresh idea you just saved.
• Parked, set aside until the timing is right.
• Resurfaced, Kosmo brings a parked idea back to your attention after 7 days, so the right moment doesn't just pass you by.
• Implemented or Archived, mark an idea done once you've acted on it, or archive it if it's no longer relevant.

Kosmo doesn't just generate generic prompts — it optimizes output for each AI tool's specific strengths and format preferences:

If you don't specify a tool, Kosmo infers the best match from your request and tells you which tool it selected.

The tools listed above are the ones Kosmo formats output for directly. Behind that, Kosmo's knowledge vault holds current, official documentation across a much wider range of frameworks, providers, and standards, so a compile can ground itself in the right facts no matter what you're building with. This list grows as the vault does.

Long-Term Memory is not available yet — it's in active development. This section describes what it will do once it ships.

Once live, Kosmo will remember context across sessions so you don't have to repeat yourself: your tech stack preferences, style patterns, and past decisions will carry forward into future compiles automatically. Until then, nothing is persisted or recalled between sessions beyond your saved Idea Library entries.

Loop Engineering is how Kosmo forces AI tools to think before they act. Instead of generating a single-pass response, the compiled prompt instructs the AI to run multi-step reasoning loops.

What is a reasoning loop?

A reasoning loop is a structured sequence embedded in the prompt that tells the AI: "First analyze the problem, then plan your approach, then implement, then review your implementation for errors, then refine." This prevents the common failure mode where AI tools produce plausible-looking but broken output on the first attempt.

Which compiles trigger loops?

• Complex code generation — Multi-file architectures, database schemas, API integrations
• Debugging requests — Requires analyzing symptoms, forming hypotheses, and verifying fixes
• System design — Architecture decisions that need trade-off analysis
• Refactoring — Must preserve behavior while improving structure

Simple requests (formatting, renaming, one-liner code) skip the loop for faster output.

Some AI agent frameworks — like Hermes and OpenClaw— define an agent's personality and behavior in a dedicated SOUL.md file, separate from its system prompt. Kosmo has current documentation for both frameworks' conventions grounded in its knowledge base, so it can compile an accurate SOUL.md for either one — not a generic template.

Describe the agent's role, personality, and rules in plain language, and Kosmo compiles it into the structure that framework actually expects, grounded in its real, current documentation.

Kosmo ships with three carefully designed themes that adapt every surface, border, and text color across the entire interface:

Switch themes from the theme toggle in the header. Your preference is saved locally and persists across sessions.

Kosmo uses a simple credit-based system. Each compile costs exactly 1 credit, regardless of complexity.

Top-up vs subscription: Pay As You Go users can buy credit packs ($5 for 10 compiles) that expire after 30 days. Subscribers get their full allotment on each billing cycle — unused credits don't roll over. You can switch plans or cancel anytime from account settings.

Pricing is in preview and subject to change before public launch.

Kosmo keeps input handling deliberately simple:

Shortcuts work globally when no modal is open. On macOS, use ⌘ instead of Ctrl.