Every agent, native. Every session, one place.

Claude for the refactor, Codex for the long grind, Grok for the quick stuff, and whichever one still has quota at 2am. Sessions runs them all in their real interfaces, side by side, so switching models never means switching workflows.

Works with the coding agents you already run

Every agent, one list. Each harness is built for different work, so use the best one for each job. New features work the day they ship, with no wait for a model to be supported, and sessions stays out of the way: one light sidebar that idles near zero.

Parallel agents, separate worktrees. Here Codex and Claude Code work in the same repo. A second agent in the same repo gets its own worktree and branch, so edits never collide. When it is done, the agent runs sessions land: rebase onto main, run your check, push. The branch note clears once the work is on main.

One MCP process per service. Instead of every agent starting its own copy of each local MCP server, every agent connects to one shared backend per server. Seven agents and three local servers make three processes. OpenCode checks it here.

Every agent stays itself. sessions-cli does not wrap agents or restyle them. Click Grok's row and Grok opens with its own keys, modes and model. Claude Code, Codex, Cursor, OpenCode, pi, Antigravity and MiMo Code work the same way.

Try sessions now

Download for macOS