Switch tools. Keep your history.
Six months of decisions should not belong to whichever CLI you happened to install first.
The problem nobody else has
Every AI coding tool stores its transcripts in its own format, in its own directory, and reads none of the others. Move from Claude Code to Codex and your last six months stay behind — the architecture you argued through, the approach you rejected and why, the command that finally worked. You do not lose files. You lose the reasoning.
One index, five tools
chat-recall reads the transcripts each tool already writes to disk and indexes them into one place. Claude Code, Gemini CLI, OpenCode, Codex and Antigravity. Ask from inside whichever you are using now and the answer covers all of them, including the ones you have stopped using.
~/.claude Claude Code sessions, plans, tasks, history
~/.gemini Gemini CLI sessions
opencode.db OpenCode sessions and todos
~/.codex Codex sessions
~/.gemini/antigravity-cliNothing is migrated and nothing is converted. Each tool keeps writing what it always wrote; the index reads all of them.
Trying a new tool costs nothing
This is the part people notice second. Evaluating a new assistant normally means starting cold — no history, no context, re-explaining the codebase. With one shared index the new tool asks the same questions of the same memory on its first run, so a trial is a trial of the tool rather than a fresh start.
Adding a sixth
Each tool is one adapter: a file that turns that tool's transcript format into canonical events. The rest of the engine — extraction, diffing, outcomes, search — is shared. If your tool is missing, docs/ADDING_A_TOOL.md is the whole job: one new file, one line in an index, no edits anywhere else.