English · 한국어
Even on hundreds of thousands of lines of existing code, your ideas turn into shipped services — automatically.
Vibe coding promises a service the moment you describe it. So why doesn't it work inside a company? In a brownfield environment — hundreds of thousands of lines of existing code tangled with complex legacy — vibe coding is close to useless. You tell the AI to read the whole codebase and work on it, but it hallucinates constantly, and fixing the result often takes longer than doing it by hand. Meanwhile, your tokens melt away.
Try Platty, the Spec-Driven Development OS. Ideas become spec documents, and spec documents become code — automatically. The spec is written so that it never conflicts with your existing services while staying logically complete. So is the code.
Platty automatically builds an encyclopedia that understands your company the way a CTO and a CPO would. Your planning and development AI agents work from that encyclopedia. The result: spec documents and code that are logically complete and never conflict with your existing services.
Platty is proprietary software of Paradigm Shift Labs, Inc.
Static analysis coverage
Real-world validated: TypeScript/JavaScript, including monorepos · Java, including multi-module repositories
Preview: Kotlin · Python · Dart/Flutter
npm install -g @paradigmshift/platty # Install the CLI
platty install # Register the agent skillsRerun platty install whenever you want to update an existing Platty plugin to
the latest published skills. After installation or refresh, start a new Codex
or Claude Code session.
Platty will register the current repository, run static analysis, review the documentation scope, and guide you through generating the source of truth.
/platty:platty-onboarding .
$platty:platty-onboarding .
. means the repository in your AI assistant's current working directory.
Onboarding uses your current conversation language. To choose a language explicitly, append the instruction to the invocation. If the conversation language is unclear, onboarding defaults to Korean.
$platty:platty-onboarding . 한국어로 진행해줘
$platty:platty-onboarding . Continue in English.
You can also ask in plain language:
Onboard this repository with Platty.
Generating documentation for large projects: Projects with a large codebase or many documentation targets may reach the AI provider's usage limit. Subscription-based providers such as Claude Code may temporarily stop generation when the plan's usage limit is reached. Completed documents are preserved, so running the workflow again resumes from the remaining work instead of starting over.
- Detailed installation: GETTING_STARTED.md
- Usage guide — English: guide/en/usage-guide.md · 한국어: guide/ko/usage-guide.md
Have something you want to build? Just throw the idea in. Platty's planning agent understands your service better than a CPO and writes the spec document for you, automatically.
AI coding with Platty is on another level. Compared to just feeding the codebase to the model, errors drop by more than half and token usage drops to less than one-tenth. Automated coding is a given — and QA time shrinks dramatically.
- Multi-repo static analysis builds a code graph — Platty automatically draws a code graph that connects every repository across your whole service. Because it analyzes code with Platty's static-analysis technology, accuracy is far higher than LLM analysis. It understands your company's code precisely, the way a CTO does.
- The code graph is summarized into a service encyclopedia — Platty analyzes the code graph to produce a natural-language document that lets you understand the whole service systematically. Like an encyclopedia, it has a table of contents and a summary per entry. It lets AI understand your service the way a CPO would.
- Planning and development agents work from the encyclopedia — results that don't conflict with your existing legacy come out right away. In the end, vibe coding becomes possible inside the company, too.
When your code is updated, the encyclopedia updates automatically. For the concepts, see guide/en/how-platty-works.md.
During onboarding, Platty analyzes the full codebase and builds the service encyclopedia, or SOT. That first run can take time and use a meaningful amount of AI provider tokens during documentation generation. It is still worth doing.
- The system needs one full read first — Platty connects hundreds of thousands of lines of code across one or more repositories into a code graph, then summarizes that graph into a natural-language encyclopedia. It is the work a new CTO or CPO would do to understand the whole company, done once up front.
- Once built, it keeps paying back — planning and development agents no longer need to reread the entire codebase for every task. They work from the encyclopedia, which is why errors can drop by more than half and token usage can fall below one-tenth compared with feeding the raw codebase directly.
- After that, updates are lighter — when code changes, Platty updates the encyclopedia from the changed parts instead of rebuilding everything from scratch. The heavy pass is the first one.
Think of the first onboarding run as paying once to save repeatedly. If a usage limit pauses generation midway, completed documents are preserved, so rerunning the workflow resumes from the remaining work instead of starting over.
Tip
Analyzing a large project?
For large company or institutional projects, contact the Platty team. We can provide a separate API path so you can run the analysis at no cost.
npm install -g @paradigmshift/platty # Install the CLI
platty install # Register the agent skillsplatty install detects Codex and Claude Code on your PATH and installs the
ordinary platty@platty plugin into every detected runtime. To target one
runtime explicitly, use platty install --runtime codex or
platty install --runtime claude.
Rerun platty install whenever you want to refresh an existing Platty plugin
to the latest published skills. After installation or refresh, start a new
Codex or Claude Code session.
Manual fallback for Codex:
codex plugin marketplace add paradigmshift-labs/platty
codex plugin add platty@plattyManual fallback for Claude Code:
claude plugin marketplace add paradigmshift-labs/platty --scope user
claude plugin install platty@platty --scope userplatty install never installs the separate platty-mcp plugin.
If you prefer a CLI-guided setup, run platty setup from the repository you
want to analyze.
For direct HTTP MCP setup, split the responsibility by role:
Server operator:
install/use full platty plugin -> platty:platty-mcp-server-setup
MCP consumer:
install platty-mcp plugin -> platty-mcp:platty-mcp-client-setup
Supported MCP URL profiles:
local -> http://127.0.0.1:3027/api/mcp
LAN -> http://<host-ip>:3027/api/mcp
remote -> https://<context-backend-domain>/api/mcp
For remote read-only retrieval against an already configured Platty MCP server,
install the MCP-only platty-mcp plugin:
codex plugin add platty-mcp@platty/plugin install platty-mcp@platty
This repository is the public distribution surface for the Platty agent plugin — the skills that teach Codex and Claude Code how to drive the Platty CLI. It does not contain the Platty engine, CLI implementation, or backend (proprietary to Paradigm Shift Labs, Inc.).
Included full platty skills: platty:using-platty,
platty:platty-cli-router, platty:platty-onboarding, platty:platty-setup,
platty:platty-mcp-server-setup, platty:platty-static-analysis,
platty:platty-docs-target-curation, platty:platty-generated-docs,
platty:platty-sync, platty:platty-sdd-spec, platty:platty-sdd-design,
platty:platty-memory.
For remote MCP retrieval, impact analysis, memory, and SDD handoffs against an already configured Platty MCP server,
install platty-mcp, which includes platty-mcp:using-platty-mcp,
platty-mcp:platty-mcp-client-setup, platty-mcp:platty-mcp-retrieval,
platty-mcp:platty-mcp-impact-analysis, platty-mcp:platty-mcp-memory,
platty-mcp:platty-mcp-sdd-spec, and platty-mcp:platty-mcp-sdd-design.
Impact analysis converges selected specs, graph classes, cross-EPIC traversal,
repository scope, and bounded source evidence into one dossier and owns only
the final §9 appendix of the selected prd.md. The SDD request flow writes
prd.md and user_stories.md; the design flow writes
system_design.md first for review, requires explicit approval, and only then
generates readiness-classified tasks.md from that approved design.
The platty-mcp plugin remains skills-only and does not ship .mcp.json or
mcpServers, because server URLs differ by deployment.
- Node.js 20–24 (Node 25 is not supported yet).
- macOS or Linux (Windows: official support planned).
- Your own AI provider credentials for the documentation step. No login or account is required to run Platty.
Platty is proprietary software of Paradigm Shift Labs, Inc.; installation and use are governed by LICENSE.md. For licensing, billing, or feature questions, use the official Platty support channel.