ContextMeld brings Claude Code and Codex sessions, search, memories, skills and MCP into one local desktop workspace. Build reviewable context handoffs without an account, telemetry, cloud backend or automatic command execution.
I built ContextMeld because useful work kept disappearing into separate Claude Code and Codex histories. I wanted one place to find an old decision, understand the project around it and prepare only the context needed for the next session.
ContextMeld reads supported local history into SQLite and keeps the workflow explicit. It does not require an account, send telemetry, run transcript commands or automatically submit anything to an agent. Memories are selected deliberately for a handoff, skills and MCP edits are reviewable, and the demo uses synthetic data without scanning normal agent folders.
This first public release supports Claude Code and Codex session formats. Cursor, Gemini CLI, OpenCode and GitHub Copilot are detection-only until their local formats can be tested safely. I would value concrete feedback on installation, detection and the handoff workflow, especially from Windows, macOS and Linux users.