Purpose
LaunchChair is a spec-driven product and feature workflow for founders and AI builders, primarily serving a way of working in which development is carried out by coding agents (such as Codex and Claude). It organizes messy product ideas into validated research, living specs, PRDs, build cards, and work that agents can execute directly.
It includes two tracks:
- Product track: takes a new idea from validation all the way to launch.
- Feature track: for feature development on existing products, skips the new-product phase, validates requirements within an existing product, defines the PRD, and prepares work that agents can build directly, without changing the tech stack or architecture.
Main capabilities
Product validation and spec generation
- Starting from ICP pain points, substitute behavior, competitor gaps, and market wedge evidence, it conducts deep competitor analysis, substitute behavior and gap analysis, and market wedge selection.
- It turns raw ideas into ICP clarity, competitor context, substitute behavior, validation tests, and market wedge direction.
- It automatically generates PRDs and living specs, concentrating positioning, pricing, and MVP scope in a single system.
Living spec
- The living spec is the central project state behind LaunchChair. Research, choices, positioning, pricing, MVP scope, screens, data models, build board cards, landing page copy, SEO, launch assets, and agent loop state are all read from or written back to it.
- The prompt engine compiles each prompt based on the current phase, bucketing rules, allowed write paths, required JSON structures, downstream dependencies, and saved project state.
Build cards and agent loop
- Build cards are implementation units divided by feature, generated from the MVP blueprint and the living spec, carrying dependencies, acceptance criteria, prompt context, evidence, QA status, SQL operations, and fix status.
- Agent loop: reads project state, queues the next action, runs the agent, accepts structured output, applies, verifies, retries or fixes, and pauses at human checkpoints.
- Human checkpoints include SQL operations, external account authorization, manual QA notes, and more.
- Agent API and MCP: let Codex, Claude, Cursor, Hermes, Grok, Kimi, Gemini, and local agents execute the queueing, applying, verifying, retrying, fixing, and checkpoint processes.
Launch and growth
- It turns the same specStyle brings SEO, GEO, AEO, and sales.
- Converts specs into landing pages, CTA strategies, schema, llms files, launch plans, outbound sequences, campaigns, leads, and signal tracking.
- Generates SEO/GEO/AEO prompts, schema, sitemaps, robots, and llms.txt from specs.
Private Starter Boilerplate
- Creates a user's private GitHub repository from the Next.js and Supabase boilerplate, already including authentication, protected routes, UI components, and SEO scaffolding.
- Code can be inspected, edited, and deployed. Includes agent instructions and architecture documentation, as well as payments, analytics, and email integrations that agents can configure as needed.
- Includes repository-local AGENTS.md, design.md, launchchair.json, and docs.
How to Use
- Connect the agent: Provide the setup prompt to Codex or Claude Code to connect to LaunchChair, approve login in the browser, and let the agent verify the connection. You can also use the CLI: install @launchchair/agent, then run launchchair setup codex or launchchair setup claude. The CLI handles login, AGENTS.md, MCP configuration, bridge startup, and connection checks.
- Run workflows: Run the Product or Feature workflow in chat. For example, "Build a booking app for dog groomers." The agent will perform product ideation, market validation, positioning and pricing, landing page and search, and other work.
- The agent operates LaunchChair through a protected stdio bridge, using scoped tokens to resolve Product and Feature projects, read specs, claim runs, complete runs, and export phase summaries.
Use Cases
- Founders want to validate and launch a new product from scratch.
- An existing product needs to launch a new feature, and you want to validate demand, define a PRD, and get a buildable design without changing the tech stack.
- You want coding agents to work from current product facts rather than vague paragraphs.
- You need to carry product decisions all the way into SEO, GEO, AEO, and sales execution.
Limitations
- LaunchChair is not a hosted AI app builder for "generate an app from one sentence," and does not provide hosted deployment or a runtime environment.
- Agents can operate the loop, but pause at real human checkpoints such as SQL operations, external account authorization, and manual QA; this is a product boundary, not a failure.
- Coding agents or model providers may be billed separately. LaunchChair pricing is subscription-based and does not hide token usage fees within LaunchChair.
- Detailed pricing amounts for specific plans and feature differences are not yet provided.