Run Contextify from the Linux command line.

Install

The installer detects your architecture, downloads the matching binary, and sets up indexing in one step.

curl -fsSL https://contextify.sh/install.sh | sh

The installer:

• downloads the contextify binary for x86_64 or ARM64 and installs it to ~/.local/bin/
• installs the Total Recall skill for Claude Code and Codex CLI
• indexes any existing transcripts in ~/.claude/ and ~/.codex/
• sets up a systemd user timer so new transcripts keep getting indexed

The install runs as your user and needs no root. If ~/.local/bin is not on your PATH, add it, then open a new shell.

To choose the architecture yourself, download the release binary directly. Use x86_64 for most servers and desktops, or ARM64 for Raspberry Pi and AWS Graviton. Unpack it and put contextify on your PATH.

Verify the install

contextify --version
contextify status
contextify doctor

status reports the database path and how many transcripts and entries are indexed. doctor checks the binary, the skill install, and the background service. If doctor reports a missing skill, reinstall it:

contextify install-skill
contextify doctor

On Linux the database lives at ~/.local/share/contextify/contextify.db, following the XDG base directory spec. Set CONTEXTIFY_DB_PATH or pass --db to point at a different file.

Keep your history indexed

Search only sees transcripts that have been indexed. The installer indexes what is already on disk and sets up a timer for new sessions, so most of the time this runs without you. To index on demand:

contextify ingest

The installer sets up background indexing through a systemd user timer that runs every 15 minutes. Check or manage it directly:

systemctl --user status contextify.timer
journalctl --user -u contextify

If the timer is missing, install it with contextify install-service. Remove it with contextify uninstall-service. On systems without systemd, the installer falls back to a cron entry.

Search your history

Start with a distinctive phrase, task ID, branch name, file path, or error string from a past session.

contextify search "sign-in handoff" --limit 5
contextify search "token refresh race" --days 30
contextify search "cloud sync" --project . --limit 10

Search uses FTS5 with stemming, so deploy also matches "deployed" and "deploying". Hyphenated identifiers are split into separate tokens, so search "ct 1703" rather than ct-1703. Use --json for scriptable output.

Drill into a result with the entry ID from the search output:

contextify context <entry-id> --before 5 --after 15
contextify projects
contextify activity --days 7

Search from your assistant

The installer adds the Total Recall skill to Claude Code and Codex CLI. Once it is installed, ask your assistant to recall past work and it runs the search for you:

/total-recall how did we fix the auth bug?

For multi-step research, delegate to the researcher subagent. In Claude Code it is query:contextify-researcher, shipped with the skill. Codex exposes it as contextify_researcher, installed to ~/.codex/agents/. Both run several queries in a read-only sandbox and return a cited summary.

To expose search to any MCP client, run the server over stdio:

contextify serve-mcp

It serves three read-only tools: search_conversations for full-text search, list_recent for recent sessions, and get_conversation_entries for the messages in one transcript.

Sync across machines

Contextify Cloud syncs indexed history between your machines, so a Linux server and a Mac laptop can share the same searchable history. Connect from the terminal:

contextify cloud setup
contextify cloud status
contextify cloud sync

For self-hosted Cloud, point setup at your own server. Prefer the interactive prompt so the API key stays out of your shell history:

contextify cloud setup --url https://your-server.example.com
contextify cloud projects
contextify cloud sync

cloud projects manages which projects sync from this machine. cloud search queries entries across every synced device, not just this one.

What runs on Linux

Related docs