Glyphox
Describe an app. Glyphox's Assistant plans it and builds it as a model in your repository, and you export runnable
source code — a React web app, a Node.js and Prisma backend, React Native, Docker and CI — all inside VS Code.
The app is stored in your repository as .glyphox/: JSON and Markdown you commit, diff and review like any other
code. The visual editors, the Assistant and every code generator read that one model, so what you design is what you
export.
Get started
- Open a folder (an empty one, or an existing repository).
- Open the Glyphox view in the activity bar and sign in to your Glyphox account.
- Select Initialise Workspace. Start blank, or from one of two samples: Acme Service Desk (tickets, a
dashboard, an escalation workflow with approvals, AI agents) or Acme Commerce Suite (four modules: two services
that own their data, a shared UI kit and a back office).
- Press F5 to run the app on your computer at
http://localhost:13000, with its Console at /console (sign in
as admin@local / admin). The first run installs the app's npm packages.
- Ask the Assistant: open the Chat view and type
@glyphox /plan followed by what you want.
Choose how the Assistant reaches a model
Glyphox does not include a model or charge for one. Use one of these:
- The model you already use in VS Code. By default the Assistant uses the model selected in the Chat view, for
example GitHub Copilot's. No setup is needed if Copilot is signed in.
- Your own API key. Anthropic, OpenAI, Azure OpenAI, Google Gemini, Mistral, Groq, OpenRouter, Ollama (local
or cloud), LM Studio and more. Run Glyphox: Select Assistant Model and select the provider and the model; the
key is stored in VS Code's secret storage, never in your repository. Glyphox: Select Planning Model sets a
separate model for the planning stages.
- A coding agent. Claude Code, Cursor, Codex, Gemini CLI and others use the same tools over MCP. Glyphox writes
each agent's project configuration when it finds the agent on your computer; Glyphox: Connect Coding Agents
adds one manually.
Without a model, the Assistant says so when you ask it something. Everything else in Glyphox works without one. The
two samples' runtime agents are set up for a local Ollama.
What you build with
- Pages on a visual canvas, composed from shadcn/ui parts with Tailwind styling, a live preview beside it,
and an app shell (sidebar, top bar) around every page.
- Logic as flows on a canvas: records, queries, branches, loops, REST calls, email, Excel and PDF documents,
AI agents, and SQL where you need it. Server actions can run on a schedule as scheduled jobs.
- Data in an ERD: tables, relationships, lookup tables, and who may read and write each one.
- Workflows for long-running processes: approvals, human tasks, timers and service calls, with a task inbox.
- Agents that run inside your app, connected to its actions, data, APIs and MCP servers.
- Code: a React component or a whole page as a
.tsx file, placed on the canvas like any other part.
- Modules: services that own their data, libraries shared between them, and ui modules for each audience.
How the Assistant works
In the Chat view, @glyphox takes a change through Requirements → Design → Tasks → Build. Each stage is a
document in your repository that you approve, or ask it to change, before the next begins. Then it builds task by
task and checks each one: validation, rendering, UX rules and the task's own acceptance tests, which it runs against
the app.
| Command |
What it does |
/plan <request> |
Plan a change and take it through requirements, design, tasks and build |
/next |
Run the next stage of the approved plan (/next all runs every stage) |
/answer |
Answer the open questions in the requirements |
/ask <question> |
Ask about the app and its record without changing anything |
/build <request> |
Build the request now, without a plan or a record |
/incident |
Write a post-mortem, then plan the change that prevents a repeat |
You can also send a plain message: the Assistant tells a question from a request, and asks before it starts a
change. The mode (Manual, Plan, Auto or Autonomous) sets how often it stops to ask.
Export and publish
- Glyphox: Export Solution writes the whole solution as source code: one service per module, a gateway, an
identity service,
docker-compose.yml, CI for GitHub or GitLab, and a Console for users, roles, jobs, workflows
and settings.
- Glyphox: Publish Solution generates the same code into a folder of this repository and stages it in Source
Control, so one commit carries the model and the code generated from it.
Export and Publish require a Pro plan or the free trial. Running the app on your computer does not.
Generated apps include security headers, an allow-listed CORS policy, rate-limited sign-in, role checks on every
table and action, parameterised SQL, sealed secrets per environment, and no default signing key.
Requirements
- VS Code 1.104 or later.
- Node.js 24.22.1 or later, to run the app on your computer.
- Docker, only to run the exported
docker-compose.yml.
Learn more
| |
