qllm is a QuickClaude-inspired launcher for AI coding CLIs.

agent-crew-style installer:

curl -s https://raw.githubusercontent.com/woogiekim/qllm/main/install.sh | bash

This installs qllm into ~/.qllm/app and creates a stable shell entrypoint at ~/.local/bin/qllm.

Local checkout install:

./install.sh

npm global install is also supported:

npm install -g qllm

After install:

qllm

The flow is intentionally short:

1. Run qllm
2. Select an AI provider
3. Select a recently used project
4. qllm launches the provider in that project

Built-in providers:

• claude -> claude
• codex -> codex
• gemini -> gemini
• cursor -> cursor

qllm guarantees provider-specific argument compatibility by treating everything after -- as opaque AI CLI arguments. Those arguments are passed to the selected provider command unchanged.

qllm --ai codex -- --model gpt-5

This is the compatibility contract for existing AI options:

• qllm options go before --: --ai, --project, --dry-run
• AI options go after --
• options after -- are not parsed, rewritten, validated, reordered, or filtered by qllm
• this applies to claude, codex, gemini, and cursor

Examples:

qllm --ai claude -- --resume --model sonnet
qllm --ai codex -- --model gpt-5 --ask-for-approval never
qllm --ai gemini -- --model gemini-pro
qllm --ai cursor -- --reuse-window --goto src/index.ts:1

Provider bypass mode launches the selected AI with that provider's own bypass option.

qllm bypass
qllm bypass --ai codex --project /path/to/project

Known provider bypass options:

• claude -> --dangerously-skip-permissions
• codex -> --dangerously-bypass-approvals-and-sandbox
• gemini -> --yolo

cursor does not currently define a qllm provider bypass option. Raw provider options are still supported after --:

qllm --ai cursor --project /path/to/project -- --reuse-window

qllm reads:

• provider project stores such as ~/.claude/projects
• qllm launch history at ~/.qllm/history.json

If no recent projects exist yet, seed history explicitly:

qllm --ai codex --project /path/to/project