deja-vu, created by Vshulcz, is an AI-search utility that gives coding assistants persistent memory for development workflows. It captures interactions from AI tools and exposes that history to Model Context Protocol clients so agents can reference past work across sessions. The app presents searchable historical context with ranked retrieval and an MCP integration point. It targets software developers, DevOps engineers, and power users who use multiple AI coding agents and need session continuity.

What tasks can you actually use it for?

The tool centralizes outputs produced by popular coding assistants and CLI agents, making prior exchanges available for programmatic queries. It accepts session data from interfaces such as Claude Code, Cursor, Gemini CLI, Codex, Aider, OpenCode, and Copilot CLI, allowing agents to pull up earlier command sequences, resolved bugs, and design notes. Users can query that history from MCP-aware clients so automation can reference concrete past actions.

How reliable and relevant are the search results?

Search quality is driven by a full-text engine built on SQLite FTS5 and a ranking scheme that blends BM25-style relevance with a frecency signal (frequency plus recency). That combination favors entries that both match query terms and show repeated, recent use, which helps surface useful commands and fixes rather than older, rarely used snippets. Performance targets fast retrieval through local indexing rather than remote lookups.

What are the integration and platform requirements?

The app speaks the Model Context Protocol and runs alongside any MCP-compliant client, including desktop clients and IDE integrations. Installation paths require either a Node.js or Go environment depending on the chosen method, and native operation covers macOS and Linux with Windows supported through WSL. Integration can be as simple as adding the executable as an MCP server in a client configuration, and the tool monitors common shells like Bash and Zsh for new interactions.

Does it require technical knowledge to get useful results?

Adoption asks for modest systems familiarity: configuring an MCP server, pointing clients at a local endpoint, and permitting the app to read session outputs. The architecture places data on the host machine and avoids mandatory cloud accounts, which fits users who manage their own toolchains. Automatic background syncing reduces manual indexing work, but administrators comfortable with local developer infrastructure get the most value.

Practical choice for developers who accept a local tooling trade-off

The tool is a practical option for developers and DevOps professionals who need persistent, private continuity between AI-assisted sessions. It suits teams and individuals prepared to manage a small amount of local configuration in exchange for searchable historical context. Expect a setup learning curve if you are new to MCP endpoints, and plan to treat retrieved fragments as context that still benefits from human review.