BSCODE

Author Bhagwat STotal Visits 2Visits in 7 Days 0

// preview

// description

Connect cloud or local models and run configurable coding agents from one desktop-style workspace.

BSCODE is an interactive AI coding agent interface organized around sessions, model providers, agents, and developer tools. Users can start coding sessions, choose among a broad catalog of cloud and local models, connect an API key, add a custom OpenAI-compatible or Anthropic provider, and select Build or Plan as the primary agent. The workspace also exposes specialized research and codebase subagents, MCP configuration, custom slash commands, themes, and explicit allow, ask, or deny permissions for twelve tools. A persistent status panel reports the active provider, model, endpoint, context size, agent, session, and catalog state while the main composer supports commands and file mentions.

Highlights

Multi-provider catalog

Cloud and local models in one list

Configurable agents

Build, Plan, and focused subagents

Tool permissions

Allow, ask, or deny each tool

Custom endpoints

Add compatible providers and models

How to Use

Connect a provider
Open Providers and select Connect for a built-in cloud service, or choose Add custom provider to enter a name, base URL, provider kind, API key, and model list. Local Ollama, LM Studio, and llama.cpp endpoints are also represented in the catalog.
Select a model and agent
Open Models to review providers and context sizes, then choose Use beside the desired model. Open Agents or press Tab in the composer to switch between the implementation-focused Build agent and the restricted Plan agent; focused subagents can be referenced from prompts.
Set tool permissions
Open Tools to review the shell, file, search, patch, task, web, question, skill, and language-server capabilities. Assign allow, ask, or deny to each tool and save the permission set before running an agent on a project.
Start a coding session
Choose New session, type a request in the central composer, and use slash commands or file mentions when relevant. The right status panel shows the active route and session state; sending model-backed work requires a connected API key or an available local model endpoint.