Module Development
Voxta Utilities: Module Development lets an assistant build the Voxta module in a workspace, load it into this Voxta, reload it after each change and read why it failed.
Voxta Utilities: Module Development lets an assistant write a Voxta module with you. In a chat working in a module's repository, it builds the module, loads it into the running Voxta, reloads it after each change, and reads the load errors and server log when it fails. No restart.
It does only what needs Voxta. Editing files is the Workspace tools' job. Git and dotnet commands go through the shell tool of Assistant Tools.
Experimental. Loading a module builds it and runs its code inside Voxta, with Voxta's access to your machine. Only load code you trust. It only works for an administrator whose account has a password, on a Voxta that loads custom modules (not Lite or hosted servers).
Setup
- Add Modules → Augmentations → Voxta Utilities: Module Development → Install (administrators only). It installs on its defaults.
- Work › Workspaces → Add a workspace, New project, Start from: Voxta module. Voxta clones the module template under the name you give, renamed everywhere, with a fresh git history.
- Under Requirements, click Allow all, and Get it on a missing .NET SDK.
- Start an Assistant chat in that workspace, with Chats may change files here on.
It turns on only where a folder asks for it: the template's .voxta/workspace.json lists module_dev, and nothing it lists runs until you Allow it for that workspace (see What a folder asks for). A chat anywhere else does not get it. An existing module repository works the same once it carries that file.
Tools
| Tool | Does |
|---|---|
load_dev_module | Builds the module in the working folder with the .NET SDK its global.json selects, then loads it as a dev folder, or reloads it. Returns the build or load errors to fix. The assistant calls it again until the module loads. Then adds the module's chat augmentations to your modules with default settings, and lists their keys. |
unload_dev_module | Unloads a module it loaded (by folder), or uninstalls an installed custom module (by id). |
test_dev_module | Runs the loaded module's own test with your settings for it, as the Test button does. A module has a test when it sets ModuleTestingProviderType. |
module_status | Lists the custom modules that loaded or failed, with their errors. Given an id or folder, the full error and stack trace. |
read_server_log | Reads the newest lines of the server log (up to 400), optionally only lines containing some text, such as the module id or ERR. Works with file logging off. |
load_dev_module runs the module's code, so it goes through the chat's tool permission prompt as destructive. It refuses a read-only workspace. A repository with several modules needs the path of one.
The chat also gets two embedded skills, loaded when needed: voxta-module-development (the edit, load, test loop) and voxta-module-anatomy (how a module's code is laid out).
Testing the module
Loading reloads the module and ends every chat using it, so test it in another chat, never the one doing the work.
- In Voxta's app, the assistant can call
open_test_chatwith the keysload_dev_modulereturned. You pick a character (the last one you chatted with is one click), and a new window starts a throwaway chat with those augmentations on. Nothing of it is saved. After the next reload ends it, the window offers Start a new test chat. Offered only in Assistant chats in the app, when Module Development is enabled. - In VaM or another app, start a chat there with those augmentation keys enabled.
A module without chat augmentations shows in Add Modules with a Dev badge: configure it there and try it in another chat.
Coding agents (MCP)
Claude Code, Codex, Copilot and other coding agents can use the same tools through MCP. The endpoint is /api/extensions/module-dev/mcp on Voxta's port. Paths are absolute: give load_dev_module the module's folder. The module template has an AGENTS.md with these steps for the agent.
- Install Module Development on an administrator account.
- Create an API key with the Module development scope on that account. Or let the agent ask for one: it calls
POST /api/device/codewith"scope": "modules.dev"and"prompt": true, and you approve the request in Voxta. - Add the endpoint to the agent with the header
Authorization: Bearer <key>. For Claude Code:claude mcp add --transport http voxta-dev http://localhost:5384/api/extensions/module-dev/mcp --header "Authorization: Bearer <key>".
The endpoint takes only a key with that scope. It refuses requests from other machines and keys of accounts that are not administrators. A signed-in browser cannot call it. The agent cannot open test chats. Start a chat in Voxta with the augmentation keys that load_dev_module returns.
The agent also gets the skills, written for the running Voxta: the module development skills, and the skills of the built-in Voxta memory book (contexts, flags, actions, scenarios and more). It reads them with the read_skill tool, or as the MCP resources skill://voxta/<name>.
Restarts
A module that cannot reload in place (it sets AllowReload = false, or replaces a built-in module) needs a restart, and load_dev_module says so. Restart from the banner on Local Development Modules (reached from Settings › AI Setup › Module Sources), or on the module's source page.
Service settings
| Setting | Notes |
|---|---|
| Build Timeout (seconds) | How long a build before loading may take. Default 300, at least 30. The first build restores packages and takes longest. |
Notes
- The workspace's requirements list also checks the .NET SDK each module's
global.jsonasks for, with where to get it. - Building and loading need the module to be a folder with
module.jsonbeside its.csproj. - Releasing the module: Publishing a module.
Workspace
Voxta Utilities: Workspace lets an assistant read and write real files in the folder a chat works in, shown as a tree beside the chat. New projects, requirements and workspace.json.
Model Context Protocol (HTTP/SSE)
Voxta Utilities. Connect to MCP tool servers over HTTP/SSE so the AI can call external tools.