frenchie
Frenchie - your agent's best friend.
Frenchie Method gives your agents a production-grade workflow from requirement to release. Frenchie Kit gives them the file tools to read, listen, extract, and create.
$result: .frenchie/project-prd/result.md > /requirement artifact: 01-requirement.md > /test-case artifact: 03-test-case.md > /implement artifact: 04-implementation-plan.md agent reads the PRD, plans the worktree, runs tests
product shape
The workflow layer for building software with agents.
Turn a rough idea into shared artifacts, test cases, implementation plans, reviews, walkthroughs, and release checks. Frenchie Method keeps the agent, the human, and the repo aligned from requirement to release.
The file and multimodal tools your agent can call.
Give your agent OCR, transcription, extraction, and generation through one Kit. Send files in, get clean Markdown or generated assets back, and keep the workflow moving. Under the hood, Kit ships as MCP tools and slash commands.
the loop
Kit handles the file context. Method turns that context into artifacts, worktree execution, and release evidence the human can trust.
Files arrive
PRDs, recordings, screenshots, spreadsheets, and docs enter the workflow as source context.
Kit makes Markdown
Frenchie Kit converts the files into clean Markdown so the agent does not guess from broken context.
Method writes artifacts
Frenchie Method turns that context into requirements, test cases, and an implementation plan.
Agent works isolated
The implementation happens in a worktree, with the artifact trail beside the code.
Review and ship
Tests, review, and release-check create the evidence a human needs before the next loop starts.
method deep dive
Frenchie Method writes the artifacts your team and agents need to continue work across sessions, worktrees, and handoffs.
Collect user intent until the agent is no longer guessing. Capture scope, edge cases, constraints, and open questions.
01-requirement.mdTurns the requirement into clear behavior, edge cases, constraints, and design guidance when the change is user-facing.
02-acceptance-criteria.mdComes before the implementation plan. Acceptance criteria are not test cases.
03-test-case.mdReads requirements, design, and test cases, then adapts the work to the project architecture and tools.
04-implementation-plan.mdReviews the actual result first, then checks correctness, tests, security, and artifact drift.
08-review-report.mdLists what is ready, risky, and missing before release. It does not merge or release.
09-release-checklist.mdkit quick scan
Frenchie Kit handles file parsing and generation outside the agent context, then returns clean Markdown or files the agent can use. It is delivered through MCP tools and slash commands.
PDF/image to Markdown for scanned specs, invoices, forms, screenshots, contracts, and handwritten notes.
Audio/video to Markdown for meetings, demos, sales calls, interviews, and user research.
Office and spreadsheet files to Markdown for DOCX, XLSX, CSV, TSV, and PPTX workflows.
Prompt to image file for visual concepts, covers, mockups, and product assets.
Stdio MCP for local agent tools. HTTP MCP for hosted and web agents.
Credit-based pricing, 100 free credits on first signup, files processed and deleted, results expire after delivery.
quick start
The plain install sets up Frenchie Method skills for text workflows. Method text workflows do not require an API key. Kit commands can be used standalone, but Kit/MCP file tools need an API key and credits when files or generated assets are processed.
$ npx @lab94/frenchie install
This sets up Frenchie Method skills and the local commands your agent can use for text-first workflow guidance.
> /frenchie-method > /prepare-feature "Describe the feature"
Method turns the request into artifacts the agent can follow and the human can review.
$ npx @lab94/frenchie login # browser approval creates and writes the CLI key > /ocr ./spec.pdf > /transcribe ./interview.mp4
Use Kit commands directly when you only need OCR, transcription, extraction, or generation. Kit/MCP file tools need a logged-in API key and credits for processing; they do not require entering the full Method workflow.
use cases
Turn an idea into requirements, test cases, an implementation plan, and release-check artifacts that survive across sessions.
Keep agents, product decisions, design notes, tests, and PR review in shared .frenchie/ artifacts.
Use /systematic-debug to investigate, then keep enough evidence to review and release safely.
Parse PRDs, recordings, screenshots, spreadsheets, and docs through MCP, then feed the context into Method artifacts.
trust, pricing, boundaries
Frenchie is not trying to replace your repo, your issue tracker, or your agent. It gives the agent a method to follow and tools to handle files. The rest stays in the workflow your team already uses.
Frenchie is not a document intelligence platform, RAG system, or project management suite.
method guide
Start with the public principles page to understand the workflow model, artifact trail, quality checks, and human approval gates. The Method articles unpack how review trails, test-first planning, and worktrees behave in real projects.
faq
Frenchie Method is a structured workflow for AI-assisted product development. It helps agents and humans move from requirement to release through shared artifacts, test cases, implementation plans, reviews, and release checks.
Method is the workflow layer. Kit is the file and multimodal capability layer. Method keeps the work traceable, while Kit turns files and prompts into usable context or assets.
No. Method text workflows do not require an API key. Hosted Kit processing, such as OCR, transcription, extraction, and generation, requires credits and an API key when those capabilities are used.
Yes. Method is useful even when you are the whole product team because it keeps the work recoverable across sessions, branches, and agent handoffs.
Yes. Teams can use Method artifacts as shared context for product decisions, test evidence, walkthroughs, review reports, and release readiness.
Frenchie is built for agents that can install skill commands and MCP tools. The installer configures supported local agents automatically and documents the hosted HTTP path for web agents.
Feature artifacts live under .frenchie/docs/feature/[feature-name]/ so another agent, teammate, or later session can continue from the same trail.
That is your choice. Teams often commit durable product artifacts and ignore local state or generated scratch files. Frenchie keeps the structure clear either way.
Worktrees keep parallel agent tasks isolated, preserve the main checkout, and make setup and cleanup explicit.
Frenchie review calls out artifact drift. The human can update the artifacts, change the implementation back, or treat the difference as a scope change and rerun the affected steps.
Kit includes OCR to Markdown, transcription to Markdown, file extraction to Markdown, image generation, async result polling, hosted upload support, and result file fetching where needed.
Files are processed and deleted. Results expire after delivery instead of becoming long-term storage.
Frenchie uses credit-based pricing with no subscriptions. You get 100 free credits on signup, credits do not expire, and hosted Kit capabilities consume credits only when used.
Yes. Kit commands can be used standalone after install. You can call /ocr, /transcribe, extraction, or generation commands without entering the full Method workflow. Kit/MCP file tools need an API key and credits when they process files or generated assets.
Install Frenchie Method for the workflow. Add Frenchie Kit whenever the work includes files.