Apphouz Craftingtable
An offline-first, spec-driven desktop design table — for software and infrastructure planning.
Electron + React shell, a Go sidecar engine, a .craftingtable/ JSON/Markdown
Single-Source-of-Truth under Git, a deterministic Work Package generator, and two output
channels: Export & Handoff (for humans) and a built-in MCP server (for AI coding
agents).
Craftingtable is first a design tool, second an AI context source. If you never launch an AI agent, the product is still fully valuable — the MCP server is just one of the output channels.
In one sentence
You plan your next build precisely — business cases, user stories, flowcharts, file structure, data models, APIs, infrastructure, design elements and technology standards — and the app assembles, without AI, deterministically, the atomic work packages that a human can pick up and work through exactly like an AI coding agent can.
What we do NOT build: not a code generator, not an AI chat, not a cloud SaaS, not a project-management tool, not a Miro clone.
The problem it solves
Primary: plan and reality drifting apart
Product knowledge scatters — in people's heads, on Miro boards, in stale Confluence pages, Word docs, Slack threads and the code. The measurable consequences: slow onboarding, gut-feel estimates, untrackable change management, audit documentation produced painfully after the fact, and "just the code" handed to the client. None of these are AI problems.
Secondary: AI context fragmentation
When a developer hands a 50+ file project to an LLM, the agent doesn't see the system-level architecture, doesn't know where a file should be, doesn't know the business intent, and doesn't know which data model, API and design token a task touches. That's where "vibe coding" comes from.
The same solution for both
Craftingtable turns intent into machine-readable, versioned artifacts, and cuts out of them exactly — and only — the context a given task needs. That package is the same whether a human reads it or an agent loads it.
The product's strongest message: "Plan it properly — and if you work with AI, it will understand it precisely too." Not the other way around.
Who it is for
| Segment | What they gain | Uses AI? |
|---|---|---|
| Companies with their own tech product | A single source of truth; onboarding, estimation, change management | Varies |
| Software houses, agencies | A handoverable spec for the client; repeatable project kick-off | Increasingly |
| Power users, solo developers | Architecture discipline even alone; faster AI development | Yes |
| DevOps / SysOps / system integrators | Infrastructure blueprint, topology, IaC plan, change log | Rarely |
| Business side (BA, PO, CTO) | Level 1–2: business case, story, acceptance criteria, report | No |
Technical essence: three things at once
- A structured editor (the surface). A desktop app where you plan along modules. Every module's output is typed JSON + companion Markdown, and the modules are forced to interconnect — this reference chain (traceability graph) is the core of the product.
- A file format and storage convention (the data). A
.craftingtable/folder at the project root, deterministically formatted JSON and Markdown, under Git. This is not the app's internal database — it is the product's output. - Two output channels (the integration). The deterministic engine produces a single artifact — the Work Package — consumable both by humans (MD/HTML/PDF) and machines (MCP). Same data, same generator, two renderings.
Where to go next
- The four-level requirement architecture — the product's conceptual backbone (L1 → L4).
- Architecture: system context — what runs where.
- Modules overview — the 11 modules and the data flow between them.
- MCP server — setting up the AI channel and its tool reference.
- Installation & platforms — supported OSes, package formats.
This doc set is currently in skeleton + core state: the architecture, module overview, MCP and installation chapters are detailed; the rest are outlines being filled in over iterations. The English translation is partial — most pages still show the Hungarian source until translated. Screenshots are placeholders.