Setup Guide

What is SindByte?

SindByte is a bridge between AI assistants (like those running in LM Studio) and your Windows computer. Normally, AI can only chat with you. With SindByte, it can:

Quick Setup in 3 Steps

Ground Truth for First Run

Connect Your AI Client

Different AI clients connect to SindByte in slightly different ways. Here are the most popular ones:

{
  "sindbyte": {
    "url": "http://127.0.0.1:5555/mcp"
  }
}

{
  "mcpServers": {
    "sindbyte": {
      "type": "streamable-http",
      "url": "http://127.0.0.1:5555/mcp"
    }
  }
}

Your First Commands

Once connected, try these natural language commands in your AI chat:

Configuration (Optional)

SindByte works immediately with basic tools. For advanced features like trading, email, or image generation, you'll need to configure API keys.

The Config Editor is the runtime control panel: feature flags, registration behavior, credentials, port, and local-only settings.

What You Can Configure

Feature Flags
Enable/disable individual tools or entire categories.

API Keys
Add keys for OpenAI, trading exchanges, email, etc.

Registration Mode
Choose how tools appear to your AI (full list vs. categories).

Security Settings
Set which tools need confirmation before running.

Advanced Features

Once you're comfortable with the basics, explore these powerful features:

Troubleshooting

1) Setup

2) Connect client

You can connect SindByte with LM Studio, Roo Code, VS Code add-ons or other MCP clients.

• Client shows the tool list -> connection is established.
• After schema changes, reload the client to clear its cache.
• Starter tool for orientation: spr_server_assistant.

3) Choose provider

• LM Studio: local, ideal for private or offline workflows.
• OpenAI: Cloud power for IQ, speech and image workflows.
• In the IQTools/Timer dialog, choose the provider via the combo box.

4) Tool filters and security

• Bit flags are used per tool: 0 disabled, 1 enabled, 2 ask first, optional +4 essential.
• The Registration Filters dialog sets per category ON, ESSENTIAL or OFF and rebuilds the published registration payload.
• Short mode can publish compact, category-centric registrations instead of the complete tool list.
• Credential-bound routes like M365 and Teams remain blocked until usable credentials are available.

Generated with the live server to make the Config Editor readable in an image: flags, registration mode, category filters, credentials and image routes.

The screenshot shows the real operation, the generated map shows the logic behind it. Together they make the Config Editor much quicker to understand than text alone.

5) Use IQTools/Timer

• Enter a prompt, choose provider/model and select the IQ checks.
• Immediate Test Prompt checks the flow immediately.
• Save the timer as once, interval, daily or weekly.
• The Result Console shows direct results for manual follow-up prompts.

6) Enable image workflows

• Set OpenAI key in the config.
• Enable OpenAIImageToolEnabled=true.
• Choose OpenAIImageModel, for example gpt-image-1.
• Then CUTools::GenerateImage and CUTools::PictureEdit are available to agents.

7) Product interface and forms

Code audit summary: the current config set defines 237 feature-flag entries in 19 MCP families. With the two core/runtime routes, the host-callable surface comes to 239 tools. By default, 121 are enabled, 76 ask first, 14 are disabled and 22 are marked as essential. Which tools a client actually sees depends on the build profile, credentials, registration filters and registration mode.

One-image summary of the server's two publishing styles.

• Full / Long publishes the enabled tool routes directly and is meant for clients that should see as many direct tools as possible.
• Short works more compactly via ToolCatalog.ListCategories, ToolCatalog.ListTools, ToolCatalog.GetToolSchema and ToolCatalog.CallTool.
• In Short mode, include_schema=true returns the full schemas. IQTools emphasizes BetterPrompt and Validate; MemoryTools emphasizes Recall and Store.
• Registration filters work per category with ON, ESSENTIAL and OFF, so the same code can publish quite differently on different machines.

8) Troubleshooting

• Tool missing: check feature flag, registration and client reload.
• OpenAI error: check API key, network and model name.
• Unstable behavior: compare the server log with the debug log and isolate the last request.
• Tests should run without MessageBox: turn off interactive messages in test builds.