Reference

Install and use the RecallBase CLI

The RecallBase CLI gives any AI agent instant access to your conversation history — across coding agents and web AI chats, all stored locally. Install it with npm, index local history into SQLite with rb import, and read it with rb today, rb search, and rb open. The browser extension is optional; connect it with rb extension install-host to add web chats.

Requirements

The CLI needs Node.js 22 or newer on macOS, Linux, or Windows. Check what you have:

node -v

1
Prints v22 or higher: skip ahead to Install and verify.
2
Says command not found, or prints an older version: download the LTS installer from nodejs.org, run it with the default options, then reopen the terminal and check node -v again.

npm comes bundled with Node.js, so there is nothing else to install, and the CLI works without a RecallBase account. The source is public under the MIT License at github.com/DarinRowe/RecallBase; npm is the verified download channel — avoid unofficial mirrors. On Linux, an install script is also available from the GitHub releases page.

Install and verify

Run:

npm install -g recallbase
rb --help

The second command must print the RecallBase command list: import, today, search, open, sources, backup, extension, extension-host, mcp. If the shell reports that rb is not found, check npm’s global binary directory and make sure it is on your PATH. Restart the terminal after fixing the path.

There is no per-command help output; rb --help is the complete command list, and the flags below are the supported surface.

Import local history

rb import
rb today
rb search "jwt retry logic"

rb import scans the supported local sources incrementally and writes its index to a local SQLite database at ~/.recallbase/recallbase.sqlite. Supported sources are OpenAI Codex CLI (codex), Claude Code (claude-code), Claude Web export files (claude-web), GitHub Copilot (copilot), and opencode (opencode). Run rb sources to see which sources were found, when each was last imported, and any diagnostics. Use rb import --source codex to restrict a run to one source.

Optional: archive web AI chats

Everything so far stands on its own — the CLI needs no browser extension, and the RecallBase browser extension needs no CLI. If you also want web AI chats, for example ChatGPT or Claude in the browser, in the same archive, connect the two. The extension captures each conversation inside the browser, and a local bridge imports it into the same SQLite archive, so one rb search — and any agent connected through rb mcp — covers both coding agents and web chats, all stored locally.

1
Install the RecallBase browser extension in Chrome or Edge — see Install the browser extension.
2
Run rb extension install-host once. It registers the local bridge with Chrome, Edge, and Firefox, then runs a health check on it. Run rb extension verify-host later to re-check without changing anything.
3
Open the extension Settings and go to Local App.
4
Click Enable local RecallBase app access, then Allow when the browser asks to communicate with cooperating native applications.
5
Click Check on the same card. The status should read Connected.

The permission is one-time; it lets the extension talk to the CLI on your machine and nothing else. From then on, captured conversations import into the local archive automatically; anything captured while the bridge was unreachable retries on its own once the CLI is back. If Check keeps reporting a problem, rb extension verify-host names the browser whose setup is missing or stale. The bridge runs on macOS, Linux, and Windows, and it is local-only: conversations move from the browser to your own database file and never to RecallBase servers.

Read the archive

Command	What it does	Useful flags
`rb today`	Summarizes today’s sessions	`--date 2026-08-10` for another day
`rb search "query"`	Full-text search across imported history, capped at 50 results	`--limit`, `--source`, `--date`
`rb open <id>`	Prints one full conversation by its id from a search result	—
`rb backup`	Writes a timestamped SQLite copy to `~/.recallbase/backups/`	`--out backup.json` for a JSON export

today and search refresh the index automatically before answering, so results stay current without a manual import. Add --json to any command for a stable machine-readable envelope, and --db <path> to point at a different database file.

Use the CLI from an agent

Two integrations run locally; pick the one your agent supports. Neither sends history anywhere.

An Agent Skill package teaches a skill-aware agent how to drive rb directly, and a global install works across projects:

npx skills add DarinRowe/RecallBase --global

Update it later with npx skills update recallbase --global.

Or point your agent at the local MCP server: rb mcp starts one over stdio with four tools — today, search, open, and sources — so a coding agent can search your local history on your machine.

{ "mcpServers": { "recallbase": { "command": "rb", "args": ["mcp"] } } }

This local server is different from the public RecallBase Docs MCP at recallbase.net, which serves product documentation only and never touches user conversations.

Data boundary

The RecallBase CLI has no cloud sync, cloud backup, cloud search, or account upload. Raw session transcripts stay on your machine unless you deliberately copy or export them with another tool. The website never receives local paths, filenames, prompts, answers, tokens, cookies, or the SQLite database.

Keep normal backups of local data you cannot replace. A local-only product avoids server-side transcript storage, but it cannot protect your machine from disk failure, deletion, or an operating-system reset.

Troubleshoot the install

If installation fails, record the npm error, operating system, CPU architecture, Node and npm versions, and the output of npm config get prefix. Do not include a session transcript or the local database. If rb --help works but an import does not, run rb sources and check the public repository documentation for that source’s current support boundary.