Introducing the Warp Agent CLI: A CLI coding agent that does what others can't. - Full access to Warp's auto routers and SOTA model selection - Full shell access (aka you can `cd` without closing your agent session) - Orchestration and cloud handoff capabilities

The CLI has a unique mux’ing architecture. Within an agent session, we are managing pty connections between the agent and the shell, a similar architecture to how tmux works. This means full shell access for you and your agent: `cd`, git commands, REPLs... even other TUIs.

The Warp Agent CLI is fully interactive, with rich UI for orchestrating and monitoring subagents.

The CLI supports any model you could want, with options for BYOK and SuperGrok subscriptions.

Routing sends requests to the right model automatically. Choose from Warp's auto routers, or configure your own custom router with the models and triggers you want.

Install it and tell us what you think: Mac / Linux (Terminal): curl -fsSL | bash Windows (PowerShell): Invoke-RestMethod " | Invoke-Expression

let’s goo @vikvang1 @varoonkodithala !!

Oh shoot this coool

Fire

Is it open source though?

Fully open source:

Seriously impressive work team 🤩

wow

sick, i need to learn more about these routers

Cool!

@andrewdsouza this could be a strong fit for the engineering teams I know experimenting with agentic workflows. Happy to see who’d be useful to meet.

Any benefits to running it in Warp instead of other terminals?

In warp it’s prob better to just use our built in agent

woahhh

the shell staying open across `cd` is the convenience, and it's also where state quietly accumulates that never shows up in a diff - env vars set three steps ago, a background process from an earlier command. powerful in the moment, harder to reconstruct why a later command behaved differently once the session's been running a while.

ok im ready to explode🤯

Seems a great job . An harness in python with ollama , it accept cd ls mkdir in the prompt without llm interpretation

Can I connect it to codex ?

Live shell access inside the agent session makes debugging feel local. The tmux style mux keeps context alive across commands. Show diff before apply

Finally, someone built a CLI agent that doesn't fight the shell. Full PTY support, persistent sessions with `cd` that actually sticks, and working TUIs. This feels like the missing piece. The mux architecture seems obvious in hindsight. Trying it tonight. Can't wait to see how the auto-routing handles real multi-step workflows.

The CLI is becoming the natural home for agents because execution matters as much as intelligence.

@vikvang1 would I need a BAA to try this out in a PHI-safe/HIPAA compliant manner? Want to give this spin.

whoaaaa

After installing it, run `warp` in a new terminal windows, it opens Warp application

this is dope can't wait to play with it!

So cool

Hey Zach, I’m doing a research project where I’m speaking with SaaS builders about how they actually went from building → getting customers → generating revenue. I’m documenting the lessons publicly on X and would love to include your experience. Would you be open to a call?

Full shell access without closing the session, that's a real quality of life fix

This is sick

Any plans to integrate with T3Code?

Add ChatGPT pro auth as an option please

🔥

CLI agents with real shell access are the right shape. The part I still want is a hard stop before the change lands so you read the diff, not the model summary.

this is super big, looking forward to token maxxing w this

does it persist state across sessions, or does it start over every time you cd into a new directory?

So what’s the difference between this and other coding agent cli?

Congrats the launch! With Jarvish (AI-native Rust shell) that I made it, REPL itself is the AI, so no "agent session" is needed. You can get native ⁠ai⁠ pipes & auto ⁠stderr⁠ capture out of the box. Check it out, please!!