// features
Everything in the toolbox, and what each part is for.
The full reference. Versioning is the foundation, Loop is the workbench, and one library feeds your terminal, browser, and agents.
Version control, for real
Every save is a version. Nothing is ever lost to an overwrite again.
Every edit becomes a numbered version with a changelog. Semantic bumps (minor/major) when a change matters.
Compare any two versions side by side, inline or unified. See exactly what changed between v2 and v3.
New version made things worse? Roll back instantly. Linear history keeps the story readable.
Share /p/your-prompt?v=3 and it stays v3 forever. Cite the exact prompt that produced a result.
Loop Engineering: your iteration workbench
“Improve” buttons are one-shot and subjective. Loop is measurable iteration against your own agent.
Generate concise, structured, and step-by-step rewrites of a prompt in one pass, grounded in the Pattern Library rather than random mutation.
ple loop --exec 'claude -p {}' runs every variant through your real agent command and shows the outputs side by side.
Pick the best variant and it lands in history with a Loop(strategy) changelog: diffable, reversible, auditable.
See the token delta (for example −32%) next to a variant before you commit to it. Same quality for less spend is a win Loop makes visible.
One library, every surface
Claude Code now pulls your claude.ai skills down, one way, but Skills still don’t sync between every surface. Your prompts should follow you everywhere.
npm i -g promptlike. ple show, copy, get --var give raw, agent-readable output. Pipe prompts straight into any tool.
Edit, version, organize, share. Your prompt repository in the browser, with the full history UI.
Your library inside Claude Code and other MCP clients. In Claude Code your prompts appear in the / menu, and agents can list, search, fill and — when you ask — save or version a prompt. Every write is a new version you can roll back; nothing is ever deleted.
Your library in a side panel next to ChatGPT, Claude, and Gemini. Fill {{vars}} and copy without switching tabs.
Better prompts while you write
The editor works like an engineering tool: structure, patterns, and cost are visible as you type.
Placeholders turn a prompt into a reusable template. Fill-and-copy on the web, --var flags in the CLI, fill_variables over MCP.
10 patterns (Persona, Few-shot, Chain-of-Thought, Output contract, Critic pass and more) as copy-ready templates.
Token count and per-call cost as you write, priced across Claude, GPT-4o, and Gemini. Know what a prompt costs before you ship it.
Structured critique that rewrites role, context, and instruction sections, and asks clarifying questions when your intent is ambiguous.
Built for daily reuse
A library you stop visiting is a graveyard. These are the reasons you come back.
Jump to any prompt by name from anywhere in the app. Keyboard-first, like your editor.
Your working set surfaces itself. Usage counts include copies from the CLI, not just the web.
Repository-style organization: projects like repos, tags for cross-cutting themes, search by title and description.
Share one library with up to 5 members, role-based access, and a pooled AI token budget with per-member audit.
Your data stays yours
Private by default, exportable always. No lock-in as a business model.
Publishing a project or prompt is an explicit, confirmed action, never a side effect.
Export your entire library on any plan, including Free. Import it back or take it elsewhere.
Row-level security on every table, two-layer authorization on every write, and audited sensitive flows.
Saving and organizing is free.
Pro adds unlimited history and the Loop workbench for $19/mo.