Grom: The Free VS Code AI Assistant Prioritizing Local-First Coding

A new open-source AI coding assistant is redefining how developers interact with intelligent tools inside VS Code. Grom, a free VS Code extension, combines real-time chat, inline autocomplete, and autonomous agent capabilities while prioritizing privacy and local-first workflows. Unlike many alternatives, it avoids telemetry, mandatory accounts, and persistent cloud dependencies, making it a compelling option for developers wary of third-party data handling.

A Local-First Alternative to Cloud-Based Assistants

Grom differentiates itself by defaulting to local model execution, with cloud providers like Anthropic, OpenAI, and Mistral available as optional additions. This approach ensures sensitive code remains on-device when using tools like Ollama or LM Studio, while still offering flexibility for teams that rely on proprietary models. The extension integrates seamlessly with VS Code’s sidebar, providing a unified interface for both conversational AI and agent-driven coding tasks.

Core Features: From Chat to Autonomous Agent Workflows

The extension’s feature set spans multiple workflows, catering to developers who need both precision and automation. In chat mode, it supports streaming responses with customizable system prompts and persistent memory across sessions. Users can switch to BUILD mode, where Grom autonomously reads files, writes code, searches codebases, and executes terminal commands—all with granular approval controls.

Key capabilities include:

• Built-in file tools: Read, write, delete, search, and list directories directly from the interface.
• Agentic safety: Every destructive action (e.g., file deletion) pauses for explicit user approval before execution.
• Undo functionality: A diff-aware revert system lets users undo individual or batch changes made by the agent.
• Task logging: A detailed log tracks every tool call, including arguments and results, for transparency.
• Retrieval-Augmented Generation (RAG): Automatically indexes the active codebase using BM25, with optional semantic embeddings via Ollama for improved context retrieval.

Inline autocomplete complements these features, offering ghost-text suggestions that adapt to typing patterns. The system slows down suggestion frequency when users rarely accept them, and supports word-by-word partial acceptance for fine-grained control.

Context Awareness and Multi-Model Support

Grom elevates context handling with @ mentions, allowing developers to attach specific elements to their queries. Supported mentions include:

• @filename – Reference a specific file in the workspace.
• @selection – Use the currently highlighted code.
• @git – Pull in uncommitted changes or diffs.
• @terminal – Reference recent terminal output.
• @problems – Include all active VS Code errors and warnings.
• @url – Fetch and attach web content dynamically.
• @docs – Search locally indexed documentation.

The extension works with a wide range of providers, from local models like Ollama and Open Code to cloud services such as Anthropic’s Claude and OpenAI’s GPT-4o. API keys are securely stored in the operating system’s keychain, avoiding exposure in configuration files.

Why This Matters for Developer Privacy and Control

The creator of Grom built the tool to address gaps in existing AI coding assistants. Most commercial options either mandate cloud subscriptions, send data to external servers by default, or offer inconsistent support for local models. By contrast, Grom’s architecture ensures local-first operation is the norm, with cloud integrations treated as optional extras. This design aligns with the needs of developers handling proprietary or sensitive code, who often prefer to keep workflows contained to their local environment.

Getting Started with Grom

Adopting Grom requires minimal setup. The process begins with installing the extension from the VS Code Marketplace. For local models, users can pair Grom with Ollama by running a simple pull command for models like qwen2.5-coder. Once installed, the Grom panel appears in the sidebar, ready for immediate use.

For cloud-based workflows, developers can select a provider from the dropdown and input their API key, which is then stored securely. The extension’s documentation provides further guidance on configuring providers and optimizing settings for different use cases.

The Road Ahead for Grom

As an early-stage project (currently at v0.3.5), Grom is actively evolving. The developer invites feedback, bug reports, and feature requests through GitHub Discussions, fostering a collaborative approach to refinement. With its focus on privacy, local-first flexibility, and comprehensive tooling, Grom positions itself as a strong contender in the crowded AI coding assistant market.

For developers seeking an alternative that balances power with privacy, Grom offers a compelling path forward—one where control over data and workflows takes precedence.