Sverklo

A local-first developer tool that provides AI coding agents with repository evidence—files, symbols, callers, tests, dependencies, diffs, and git-pinned decisions—before they edit code, solving the problem of agents making changes without full context. It targets software developers and engineering teams using AI coding agents like Claude Code, Cursor, Windsurf, and Codex-style tools, across companies of all sizes. It is positioned as a proof-first approach, with a no-write trial mode and a public benchmark, and is delivered as an open-source, MIT-licensed CLI and MCP server that runs locally by default with telemetry off.

Key features

• Run proof before agent edits
• No-write trial mode
• Guided symbol selection
• Repo memory proof with caller graph
• Markdown receipt generation
• Impact analysis for signature changes
• Context task search with scores
• Bundle files and memories
• Review diff for branch changes
• Test map with coverage and risk
• Global MCP config for all repos
• Bi-temporal memory pinned to git SHAs
• Hybrid retrieval with RRF scoring
• Public 180-task benchmark