Forums

Complementing the Q&A interface, Forums offers a command-line tool called remote-bash that allows developers to execute arbitrary bash commands against any public GitHub repository without a local clone. This CLI, invoked via `npx remote-bash <repo>`, supports common operations like grep to search for exports, find to locate files, and ls to list directories. It accepts flags to target a specific branch (`--ref`) or version tag (`-v`), enabling precise inspection of different release states. For instance, running `npx remote-bash next -v 15.0.0 -- ls -la packages/` reveals the structure of a particular Next.js version. The tool runs in a sandboxed, read-only environment, so you can safely explore without risk. Developers can also add it as a skill via `npx skills add basehub-ai/forums --skill remote-bash`, integrating it into their workflow for quick, scriptable code exploration. This reduces the need to download massive repos just to check a single detail.

Forums extends its capabilities to AI agents through Model Context Protocol (MCP) integration. By installing the Forums MCP, developers can allow their coding assistants—such as Claude Code, Cursor, Codex, Gemini CLI, VS Code, or Zed—to autonomously ask questions and execute bash commands on public repositories. The MCP exposes two tools: "ask" lets the agent inquire about any aspect of a repo's source code, returning source-backed answers that the agent can incorporate into its reasoning; "bash" runs read-only shell commands in a sandboxed environment, perfect for codebase exploration, running analysis tools, or checking dependencies. This agent-to-agent collaboration automates the process of understanding external code when writing programs. For example, when an agent needs to understand how a plugin system works in sst/opencode, it can use the ask tool to get a detailed, code-verified explanation, then use bash to list plugin-related files, all within a single conversation. This deep integration makes Forums an essential component of modern AI-assisted development stacks.

The core workflow of Forums revolves around a few simple steps: a user or agent poses a question about a public GitHub repository, the system clones the repo into a secure environment, explores the directory structure and file contents, and then applies a frontier LLM to analyze the source code in the context of the query. The LLM identifies the relevant code segments, constructs an explanation, and attaches source references to justify the answer. This process is fully automated and sandboxed, ensuring that no modifications occur to the original repository. The "I'm feeling lucky" button on the website offers a quick-start experience for serendipitous exploration. Whether accessed via the web, CLI, or MCP, the underlying engine consistently delivers source-backed responses. The methodology combines retrieval-augmented generation with code parsing, providing a level of accuracy that plain LLM chats cannot achieve because it grounds answers in the actual codebase, not just the model's training data.

Developers use Forums in numerous practical scenarios. When integrating a streaming AI library like vercel/ai, a developer can ask about the onFinish callback's behavior during stream aborts to avoid frontend crashes, receiving a code-linked answer that clarifies event ordering. An open-source contributor onboarding to a project like transitions.dev quickly discovers the tech stack and directory layout by asking a single question, shortening the setup time from hours to minutes. A team building on next.js uses remote-bash to grep for a specific export across different branches, ensuring the API they depend on hasn't changed. AI coding agents leverage the MCP ask tool to understand plugin architectures in sst/opencode, enabling them to suggest compliant extensions. A DevOps engineer uses remote-bash with a version tag to audit a dependency's security-related files. In each case, the outcome is faster, more accurate understanding, reduced trial-and-error, and a more efficient development cycle rooted in source truth.

Forums caters to a wide range of technical users: software developers building applications that depend on open-source libraries, open-source maintainers who need to quickly answer community questions, AI agent developers crafting autonomous coding workflows, and researchers analyzing code patterns. It supports multiple access methods: a browser-based interface for ad-hoc questions, a powerful CLI for scriptable operations, and MCP servers for integration with AI editors including Claude Code, Cursor, Gemini CLI, VS Code, and Zed. The platform is built on a scalable architecture that clones and sandboxes repositories on demand, pairing cloud-based infrastructure with advanced language models. While no pricing is mentioned, the tool currently appears free to use. In summary, Forums transforms the way developers interact with GitHub code by providing source-backed answers for GitHub repositories, making codebases instantly understandable and empowering both human and AI agents to work smarter, not harder.