🎬 New — watch the 2-minute guide videos →

← Blog

Running brcode in Warp: a coding agent in a terminal built for it

brcode is a TUI agent, and Warp is arguably the best place to run one: real blocks, great rendering, and enough workflow machinery to make the agent feel installed rather than launched. Nothing special is required — brcode is a normal CLI — but a few Warp-specific touches make it noticeably nicer. Here's the setup we use.

Zero to agent

brew install bharatrouter/tap/brcode
brcode login   # Warp hands off to the browser; the key lands back in the CLI
brcode init
brcode

That's it — the agent runs in the Warp pane like any TUI. The login loopback works exactly as in any terminal: brcode login opens your browser, and the key is caught on localhost with zero pasting. If your org already set a profile, a brand-new laptop is coding in under a minute.

A dedicated launch configuration

Warp's launch configurations restore a whole window layout in one action. A good one for agent work: pane 1 runs brcode, pane 2 is a plain shell in the same repo for git, tests, and scratch commands. Save it once (Command Palette → Save New Launch Configuration) and "open my coding agent" becomes a single command the next morning.

Warp workflows for brcode commands

Warp workflows (saved, parameterised commands) fit brcode's management surface well. The ones worth saving:

brcode doctor                 # first thing when anything misbehaves
brcode model                  # list/switch the default model
brcode usage                  # org ₹ / requests / per-agent spend
brcode identity status        # who this machine codes as + caps
brcode sandbox status         # what the agent may touch

Because Warp keeps each command's output in its own block, brcode usage and doctor outputs stay scannable and shareable — right-click a block to copy or share the whole thing with a teammate.

Keep the two AIs out of each other's way

Warp ships its own AI features; brcode is an agent with its own loop. They coexist fine with one rule: the agent owns the repo, Warp AI owns the shell. Use Warp AI for one-liners ("find which process holds port 8080"), and the brcode pane for anything that edits code. If Warp's autosuggestions distract inside the agent's TUI pane, turn them off for that pane — the agent has its own completions.

Shortcuts that matter inside the agent

brcode's engine uses ctrl+x as its leader key: ctrl+x m switches models mid-chat, ctrl+x n starts a fresh session, tab cycles subagents, escape interrupts. Warp passes all of these through cleanly. The full table is in the brcode guide.

Identity, budgets, sandbox — same as anywhere

Everything governed about brcode is terminal-agnostic: brcode identity on gives the machine its own revocable agent identity, brcode identity budget 500 caps its monthly ₹ spend, and brcode sandbox on scopes it to the project (with a macOS write-jail in strict). Warp just makes the output nicer to read.

Warp is Warp Terminal's product; brcode is not affiliated with Warp. Full reference: the brcode guide · migrating from Codex or Claude Code.