@quiq/cli

Official command-line connector for the Quiq platform — built for AI agents.

@quiq/cli (the quiq command) lets an AI agent operate Quiq on a person's behalf, over the platform's typed RPC surface. It is official Quiq software, published and maintained by Quiq.

What this is (and isn't)

• A connector, not a typical CLI. Think of it as a plugin for the Quiq platform — closer in spirit to an MCP server than to a hand-run command-line tool, but with a higher ceiling: it exposes Quiq's full typed RPC surface and adds filesystem niceties (large results spill to disk, presigned artifacts stream straight to files) that keep an agent's context lean.
• Meant for AI, on behalf of humans. The ergonomics target an autonomous agent invoking commands, not a person typing them. A human sets up and approves a session; the agent drives it.
• Safe to install and run. Apache-2.0 licensed, two runtime dependencies (commander and jq-wasm), no install or postinstall scripts, and the published tarball is nothing but compiled JavaScript you can read. Every bit of platform access is gated behind an explicit, human-approved, per-session credential — the CLI cannot reach anything a user has not granted it.

Install

npm install -g @quiq/cli

Requires Node.js ≥ 20.

Quickstart

# 1. Establish a session (the user approves it once, in the Quiq web app)
quiq session start --host <your-quiq-domain> --client <client>   # prints an approval URL
quiq session complete                                            # after the user approves
                                                                 # (no code to paste back)

# 2. Bootstrap — materialize the whole session surface to disk, then read it
quiq bootstrap --session <id>                  # (or just `quiq --session <id>`)

--client identifies the AI environment driving the CLI: one of claude-code, claude-desktop, claude-cowork, claude.ai, codex, chatgpt.com, github-copilot, other.

bootstrap is the one command that talks to the platform to fetch everything the agent needs — orientation, the typed schema, the help corpus, and the task skills — and writes it to the session's local cache. Everything after it reads those local files, no further discovery call required:

quiq schema --session <id>          # the local OpenAPI schema — pipe to `quiq jq`
quiq skills --session <id>          # list task-recipe skills; `quiq skill <id>` reads one
quiq help --session <id>            # list reference topics; `quiq help <topic>` reads one
quiq rpc <Service.Method> --session <id>   # invoke an RPC (add -h for one method's shape)

The bootstrap bundle is self-describing: it tells the agent what it can do, how to invoke it, and how results come back. This README stays deliberately thin on purpose — that guidance lives on the server and evolves without a CLI release. Read the bootstrap output before authoring anything.

How it works

• Bootstrap, then read local. A single quiq bootstrap materializes the session's orientation, schema, help, and skills to the local cache; every other verb reads those files rather than making its own discovery call. It's idempotent — re-run it to refresh from the server.
• Dynamic over the platform. Nothing is hardcoded: the surface derives entirely from the platform's live, typed schema, so new capabilities appear without a CLI update.
• Context-friendly. Large responses and artifacts are kept out of the model's context — spilled or streamed to disk for the agent to read on demand (quiq fetch pulls presigned artifacts) — rather than inlined into the transcript.
• Session-scoped and explicit. Each session carries its own bearer credential (sender-constrained with DPoP), and every command targets an explicit session — never a silent default.

License

Apache-2.0 — see LICENSE.