Flt

flt

You build an agent in Claude Code. It's good. Then you need one in Codex for the fast stuff, one in Gemini for the cheap stuff. Now you have three tools that don't know about each other, three different instruction formats, three ways to check status. You're locked into whichever ecosystem you started with, or you're managing three separate workflows.

flt makes every AI coding CLI feel like the same tool. `flt spawn`, `flt send`, `flt kill` — works the same whether the agent runs in Claude Code, Codex, Gemini CLI, OpenCode, or SWE-agent. Agents message each other across CLIs. A Claude Code orchestrator can spawn a Codex coder and a Gemini researcher, and they all report back through the same inbox.


Install

bun install -g @twaldin/flt-cli

Requires [Bun](https://bun.sh) and [tmux](https://github.com/tmux/tmux). You need at least one AI coding CLI installed (`claude`, `codex`, `gemini`, `opencode`, or SWE-agent). `git` is required for worktree-based agent isolation.

Quick start

flt init                    # initialize fleet + start controller
flt tui                     # open the terminal UI

# spawn agents in any CLI from the same interface
flt spawn coder -c claude-code -m sonnet -d ~/project "fix the login bug"
flt spawn fast -c codex -m gpt-5.3-codex -d ~/project "add input validation"
flt spawn researcher -c gemini -m gemini-2.5-flash -d ~/project "find similar auth patterns"

# talk to them
flt send coder "also add tests for the edge case"

# they talk to each other
# (from inside the researcher agent)
flt send coder "found 3 patterns in the codebase, see PR #12 for details"

# watch, kill
flt logs coder
flt kill fast

Save presets so you stop typing flags. The name is the preset — if a preset matches the agent name, it resolves automatically:

flt presets add coder -c codex -m gpt-5.3-codex
flt spawn coder -d ~/project "fi