CoDev

CoDev — AI Coding Agent Hub. Install, configure, and manage multiple AI coding agents.

Requires Node.js ≥ 22.5 (Node 24+ recommended).

Install

npm install -g codev-ai

Then run:

codev install

Commands

| Command | What it does | | ---------------------------- | ----------------------------------------------------------------------------------------- | | codev --help, -h | Show help | | codev --version, -v | Show version | | codev install | Install and configure AI coding agents | | codev update | Update installed AI coding agents | | codev export | Export this directory's agent conversations to ~/.codev/logs/<project>/ | | codev upload | Export, then upload conversation logs to the backend | | codev claude | Run the claude CLI (forwards remaining arguments) | | codev claude --restore | Restore ~/.claude/settings.json from ~/.claude/settings.json.backup | | codev codex | Run the codex CLI (forwards remaining arguments) | | codev codex --restore | Restore ~/.codex/config.toml from ~/.codex/config.toml.backup | | codev opencode | Run the opencode CLI (forwards remaining arguments) | | codev opencode --restore | Restore ~/.config/opencode/opencode.json from ~/.config/opencode/opencode.json.backup | | codev logout | Sign out of SSO |

Restoring a previous configuration

CoDev will replace ~/.claude/settings.json, ~/.codex/config.toml, and ~/.config/opencode/opencode.json with new configs. Before writing its own config, CoDev backs up the specific file it would replace — other files in those directories are left untouched.

| Selection | Backed up | | ----------- | ----------------------------------------- | | Claude Code | ~/.claude/settings.json.backup | | Codex | ~/.codex/config.toml.backup | | OpenCode | ~/.config/opencode/opencode.json.backup |

settings.json, config.toml, and opencode.json are replaced (not merged), so any keys you had before live only in the file backup.

Existing backups

If a backup already exists from a prior CoDev run (*.backup), CoDev leaves it untouched and proceeds to replace the live config. The existing backup is assumed to be your pre-CoDev original and is never clobbered by later runs. To capture a fresh backup, delete the old *.backup first, then run codev install again.

Restore

Use the built-in restore shortcut:

codev claude --restore
codev codex --restore
codev opencode --restore

Each command removes the active config file and renames the corresponding *.backup back into place. If no backup exists, the command prints a "No backup found" message and exits with code 1.

Or do it manually:

# Claude Code
mv ~/.claude/settings.json.backup ~/.claude/settings.json

# Codex
mv ~/.codex/config.toml.backup ~/.codex/config.toml

# OpenCode
mv ~/.config/opencode/opencode.json.backup ~/.config/opencode/opencode.json

If you have a session running, you might need to restart it with claude -c, codex resume, or opencode -c to resume your progress.

Exporting conversation history

codev export reads each agent's on-disk session store, filters to conversations that belong to the current directory, and writes them as Markdown to ~/.codev/logs/<project>/<agent>/. Nothing is uploaded — the files stay on your machine.

~/.codev/logs/works-repos-codev/
  claude-code/
    2026-04-27_18-32-05Z-help-me-fix-the.md
  codex/
    2026-04-27_19-15-22Z-refactor-auth.md
  opencode/
    2026-04-27_20-44-10Z-explain-the-build.md
  statistics.json

• The project subfolder is the current directory's path with the home prefix stripped and non-alphanumeric characters replaced with -.
• Filenames are <UTC-timestamp>-<slug>.md, where the slug comes from the first user message in the session.
• statistics.json records per-session metadata (message counts, byte size, provider, timestamps), keyed by session ID and merged across runs.
• Existing files are overwritten on each run; sessions with no activity in the current directory are quietly skipped.

Uploading conversation history

codev upload re-runs codev export, then ships any new or changed Markdown logs from ~/.codev/logs/<project>/ to the CoDev Supabase backend. Authentication uses the same SSO login as codev install; if you're not signed in, the browser flow runs first.

• Files are SHA-256 hashed and compared against the server. Unchanged logs are skipped, so re-running is cheap.
• Each upload records the previous version it replaces, so the backend keeps history rather than overwriting.
• Payloads are gzipped over the wire.