Prompt HUD
Press Command-Shift-Space from any app, search by title or alias, and insert at the cursor.
Prompt systems and evaluation
Prompty
Bring your best prompts to the work—and prove they work.
Prompty is a Mac-native system for retrieving and transforming prompts, deploying durable agent loops, and testing prompt or model changes against evidence you can inspect.
See Prompty in action Private preview · macOS 14 or later
Search prompts or type an alias… ⌘ K
Critique a product decision
Stress-test the evidence, tradeoffs, and assumptions…
/critique-decision Turn notes into a brief
Structure raw material without inventing certainty…
/make-brief Research findings gate
Check traceability, confidence, and contradictions…
/findings-gate PROMPT
Critique a product decision
Review the decision below. Separate evidence from inference. Name the strongest counterargument.
decision_context
Copy Insert prompt
See the workflow
Retrieve. Transform. Prove.
Prompty stays close to the work: bring forward what you know, act where the text is, then test the change before it becomes part of your system.
›critique
01
Critique a product decision/critique-decision
↩ 02
Research findings gate/findings-gate
03
Audience perspective/audience-view
I think maybe the current scope is perhaps too broad.
COMMAND MODE Make this direct
The current scope is too broad.
Replace selection ↩EXPERIMENT 04 · COMPLETEOnboarding critique
Ready to ship CURRENT20%1 passing
→ DRAFT100%5 passing
4 fixed 1 held 0 regressed
The premise
Your library should come to you.
Most prompt libraries become another tab to search, copy from, and forget. Prompty lives at the point of use. Open the HUD over any app, type a few letters or an alias, and insert the prompt where your cursor already is.
When the work is selected text rather than a saved prompt, Command Mode can clean it up, translate it, apply a reusable transform, or answer a question without replacing the selection. Dictation uses the same direct path back into your work.
Command Mode
Select text and press Command-Shift-K to clean up, transform, translate, or Ask.
Hold to dictate
Hold Command-Shift-V, speak, and release. Prompty transcribes on-device, cleans, and inserts the result.
Instant capture
Press Command-Shift-S to send selected or copied text straight to your library, with a title, type, and tags suggested for you. Passwords, API keys, and one-time codes are detected and never stored.
Collections
Group related prompts and reference material into one bundle you can scope per app, hand to an agent whole, or export as a single shareable file.
Recents
One feed of what you just used, copied, transformed, or dictated — plus what your assistants reached for. Re-run any of it with a keystroke.
Beyond a prompt library
The durable layer around your AI work.
Prompty treats prompts as working infrastructure: reusable, deployable, connected to context, and testable before a change becomes part of how you work.
Agent loops
Compose goals, routines, scorers, and verifiers with explicit success checks, guardrails, and safe handoffs.
Inside Claude and Cursor
Connect once and your prompts and saved contexts appear there as tools and slash-commands. The apps talk to Prompty locally, you choose exactly what each assistant can see, and a live feed shows what they reached for.
Leverage Packs
Install, update, or remove curated role systems for product, design, engineering, research, leadership, and more.
Audience Dossiers
Keep lifecycle, evidence, hypotheses, and audience context in a durable artifact that prompts and loops can check.
Proof evaluations
Compare current and draft behavior across challenge and preservation cases, with deterministic checks or pinned model rubrics.
Runtime deployment
Install project or global skills, export goal commands, and schedule recurring work for Claude Code or Codex. Recipes keep the models, thresholds, cases, and runtime policy behind a decision reusable without losing provenance.
Guided model auditions
Find the model that clears your standard.
Build one Proof Set, then give cloud, subscription, and local models the same work. Prompty recommends only models that clear your quality bar, using consistency and speed to separate close results.
Choose models
Use the connected runtimes and local model apps you already trust.
Run one fair test
The same cases, assertions, and trials run against every candidate.
See the best fit
Quality first. Then repeatability, regressions, and speed.
Consent before execution
Know what will run before anything leaves your Mac.
Prompty shows the exact runtime and model it can verify, request count, expected duration, billing mode, and data boundary before enabling a run. Completed evidence remains inspectable even when another model fails or an audition is stopped.
Exact runtime Request budget Data boundary Evidence history
Privacy and control
Local is the default, not a paid tier.
Prompty uses Apple's on-device model by default — no API key, no cost, and your text never leaves your Mac. Your library is stored locally and speech is transcribed on-device. When you want a larger model you can bring your own provider key, point at Ollama or another local server, or connect an official subscription runtime. Prompty does not use your work to train models.
Agent access and evaluation evidence are local and inspectable. Missing variables stay visibly missing, exact model identity is preserved when a runtime can prove it, and deployed files use a never-clobber contract so an app update does not silently overwrite your edits.
Who it is for
People whose prompts became part of how they work.
Prompty is built for designers, developers, writers, researchers, operators, and leaders with a real working library spread across notes, chat histories, repos, and memory. It is for making that library usable without turning it into another inbox.
Availability
Follow the work before launch.
Join the preview or launch list. Product and release details will be shared when the app is ready.
Private preview
Prompty
Release updates
Be first to know when there is a build ready to try or a confirmed release date.
Use your email app or copy hello@labsly.com for another client. No account or payment is required.
What is included
- ✓Local prompt library, aliases, Collections, capture, and direct insertion
- ✓Command Mode transforms, hold-to-dictate, and the Recents feed
- ✓Prompts and contexts as tools and slash-commands inside Claude and Cursor
- ✓Contexts, audience dossiers, Leverage Packs, and agent loops
- ✓Proof evaluations, reusable recipes, and guided model auditions
- ✓Claude Code and Codex deployment plus local MCP access
- Access
- Preview and launch updates
- Platform
- Mac
- Delivery
- Mac App Store
- Support
- Public support policy
Product information
Support and legal.
The public URLs used for Prompty support, privacy disclosure, terms, and licensing.