DeepAI Code CLI Tools

English | 日本語 | 中文

Use DeepAI's models and the built-in Pi coding agent right from your terminal and opencode — reusing your existing DeepAI Code login and saved account settings. No API key re-entry required.

Install

curl -fsSL https://raw.githubusercontent.com/yukurust/deepai2cli/main/install.sh | sh

That is the whole install: the installer downloads just the command files into ~/.local/bin (and the /deepai command into ~/.config/opencode/command) — no repository checkout, no git. Variants:

URL=https://raw.githubusercontent.com/yukurust/deepai2cli/main/install.sh
curl -fsSL "$URL" | sh                       # install
curl -fsSL "$URL" | sh -s -- --dry-run       # show what it would do first
curl -fsSL "$URL" | sh -s -- --uninstall     # remove the installed files
curl -fsSL "$URL" -o install.sh && sh install.sh   # inspect the script before running

Re-running the curl command updates the files. To lock the version, replace main with a tag or commit in the URL. The installer checks the prerequisites, tells you when the DeepAI Code app is missing, and keeps an existing pi on your PATH (skipping the pi-side commands) so an opencode-only setup installs cleanly. Make sure ~/.local/bin is on your PATH (export PATH="$HOME/.local/bin:$PATH").

Usage

Run deepai-ready before opening opencode or pi — it is the one preparation step both need. Then use whichever commands you like:

Command	What it does
deepai-ready	Prepare: start DeepAI Code if needed, wait for its API proxy, sync opencode's config — launches nothing
pi	Run Pi as an independent session (pi "prompt" = one-shot, pi --print "…" for scripts)
deepai-pi	Friendlier wrapper around pi
deepai-opencode	deepai-ready + start opencode, in one step
deepai-up	Only start DeepAI Code and wait for its proxy (used inside deepai-ready)
/deepai …	opencode command: run the DeepAI CLI (Pi) once

deepai-ready && opencode                          # or: deepai-ready && pi
deepai-opencode                                   # the same, in one command
opencode run --model deepai/glm-5.3-flash "…"     # pick any DeepAI model in opencode
orca agent launch --agent pi --prompt "…" --json  # open an independent pi session from Orca

What each piece does internally is described in How it works.

Requirements

• macOS (the DeepAI Code app path is assumed; the built-in curl and python3 are used)
• opencode installed (for the opencode side)
• Signed in to the DeepAI Code app

How it works

DeepAI Code's local proxy

Every model call goes through a local HTTP proxy inside the DeepAI Code app. The login state lives in ~/.deepai-code/userdata/deepai/pi/deepai/models.json (providers.deepai: baseUrl + apiKey + models). Every time the app (re)starts, the proxy port and the API key rotate — which is why a previously working setup suddenly breaks with ECONNREFUSED, and why the tools re-read models.json on every run instead of caching it.

deepai-ready / deepai-up

deepai-ready is the single preparation step and launches nothing. It calls deepai-up — start the app hidden (open -g -j) if the proxy is down, then poll models.json until the proxy answers (so it follows the rotation) — and then syncs opencode's config. deepai-up is idempotent ("already running" exits immediately), so the pieces compose and never conflict; it is also available standalone. pi deliberately does none of this: if the proxy is not ready it exits with run deepai-ready first, keeping one single entry point for preparation.

opencode config sync

The sync updates only providers.deepai in ~/.config/opencode/opencode.jsonc (model list + credentials taken from models.json). Your settings, other providers and your default model are untouched; pick deepai/… models with /models or --model. JSONC comments are not preserved, and before the first change exactly one backup opencode.jsonc.bak is saved (created once, never overwritten — no backups accumulate). Credentials live only in that file (permissions 600). Model names mirror the DeepAI Code app (e.g. mimo-2.6-pro, claude-opus-5-5, glm-5.3-flash, deepseek-v4.1-flash, gpt-6.1-sol, kimi-k3), and per-model API quirks (like gpt-6-astra's Responses API) are mapped automatically. To hide other providers from the picker, add { "enabled_providers": ["deepai"] }.

pi

pi runs the Pi coding agent bundled inside the DeepAI Code app — the same engine as pi.dev — with DeepAI Code's saved account. Arguments pass straight through to the Pi CLI (pi [options] [--] [@files...] [messages...]). Each invocation is its own independent session (one JSONL file per session; --session-id / --session continue an existing one). Defaults: model deepai/mimo-2.6-pro (DEEPAI_MODEL overrides), tools read,bash,edit,write,grep,find,ls. Standalone sessions print DeepAI tools are unavailable — those app-level tools (threads, delegation) need the DeepAI Code app's local connection; the agent itself works normally.

Orca

Orca knows pi as a TUI agent (detectCmd: pi, prompt passed via argv), so orca agent launch --agent pi opens an independent pi session (a chat, or --surface terminal to run the CLI). Launch-time model selection for pi is refused by Orca ("Agent pi does not support launch-time model selection") — use opencode for model selection from Orca (opencode models feeds Orca's model picker; opencode run --model deepai/… runs them), or switch inside the session (/model deepai/…) / start with DEEPAI_MODEL=… pi from an Orca terminal. Wire deepai-ready into an Orca automation to run before opencode launches.

/deepai (opencode)

/deepai <task> runs the DeepAI CLI (Pi) once through opencode's session model; !deepai-pi runs it without involving opencode's model.

Troubleshooting: models missing or Model unavailable

1. deepai-up --status — is the local proxy alive? (run deepai-ready if not)
2. deepai-ready — re-sync the config
3. Still broken: check that baseUrl matches between ~/.deepai-code/userdata/deepai/pi/deepai/models.json and ~/.config/opencode/opencode.jsonc

Repository layout

Path	Role
pi	Runs the Pi CLI as an independent session (the entry point Orca's --agent pi uses)
deepai-pi	Friendlier wrapper around pi
deepai-up	Starts DeepAI Code in the background and waits for its local proxy
deepai-ready	Pre-flight before opencode/pi: start DeepAI Code, wait for the proxy, sync opencode's config
deepai-opencode	deepai-ready + start opencode
install.sh	Idempotent installer (downloads the commands into ~/.local/bin)
opencode/command/deepai.md	opencode /deepai command definition

License

MIT