What problem does Locus solve for AI coding tools?

Locus addresses the issue of AI coding tools losing context between sessions. It provides persistent, project-aware memory to retain architectural decisions, codebase structure, and session history across your AI development workflow.

Does Locus store sensitive information or raw file content?

By default, Locus stores metadata only. Raw file content is never written to disk unless you explicitly opt in. It also features content redaction and robust security layers to protect your data.

What is 'Carbon Copy' and how does it work?

Carbon Copy is a unique feature that passively captures conversation events – prompts, AI responses, and file changes – via an inbox-based event protocol. This ensures zero token consumption on write and processes events into searchable history for later recall.

Which AI coding clients are compatible with Locus?

Locus offers seamless compatibility with major AI coding clients, including Claude Code, Codex CLI, Cursor, Windsurf, and any other tool that supports the Model Context Protocol (MCP).

Locus: AI Coding Tools with Project-Aware Persistent Memory

AI coding tools often suffer from a lack of persistent memory, losing context of architecture decisions, file structures, and past conversations between sessions. Locus addresses this by introducing three robust memory layers: structural (an auto-parsed map of codebase elements), semantic (explicitly saved, redacted architectural decisions), and episodic (a compressed history of session activities). It further enhances memory with "Carbon Copy" for zero-cost passive capture of conversation events and provides deep integration for tools like Codex CLI, offering automatic and manual import of session history. As an MCP server, Locus seamlessly integrates with any Model Context Protocol-compatible client, offering 13 memory management tools and 3 auto-injected resources to provide a consistent, intelligent context for AI development workflows.

Key Features

01Three persistent memory layers: structural (auto-parsed codebase map), semantic (user-curated decisions), and episodic (auto-captured session history).

02Thirteen Model Context Protocol (MCP) tools for exploring, searching, remembering, and managing project memory.

03Carbon Copy: Passive conversation capture via an inbox-based event protocol, processing prompts, AI responses, and file changes without token consumption on write.

04Seamless compatibility with major AI coding clients like Claude Code, Codex CLI, Cursor, and other MCP-compatible tools.

0511 GitHub stars

06FTS5 full-text search across all memory layers and conversation events, alongside incremental scanning strategies for up-to-date context.

Use Cases

01Maintaining context and recalling past architectural decisions or recent work across AI coding sessions.

02Troubleshooting and diagnosing memory capture or integration issues for AI coding clients like Codex CLI.

03Enabling AI tools to understand project structure, imports, and exports without consuming tokens.