Find your next open-source project

Search by name, purpose, or keyword — or paste a GitHub repository URL

httpdone/magpie-aws-quick

AI Engineering

A magpie plugin that lets your own AWS Quick account serve models to coding agents.

121
19
JavaScript
MIT
new today

What it does

magpie-aws-quick is an OpenCode provider plugin for magpie that turns your own Amazon Quick (AWS Quick) sign-in into a provider named aws-quick. Agents routed through magpie — Claude Code, Codex, Gemini CLI, OpenCode — can then call the models behind Quick, and you can list the exposed models with magpie models | grep aws-quick.

The plugin does not capture packets, run a proxy, or patch the app. It reuses the Python backend bundled in the Quick desktop app (quickwork-agent), starts it locally in --outpost-only mode, reads the session the desktop app holds, and opens the app's bidirectional WebSocket chat protocol. OpenAI-style messages are translated to Quick turns and back, including streaming text, reasoning content, and usage.

A notable feature is thread reuse: the plugin tracks exact continuations with an incremental hash chain, falls back to fuzzy anchors, and finally to a gateway session key, so Quick's server-side history can be continued without resending the whole transcript. Threads are scoped by account, agent mode, thinking effort, system prompt, and tool names, and the cache stores only hashes and thread ids.

Quick start

Prepare magpie and choose a sign-in path:

  • Browser OAuth: no desktop app required. During magpie login aws-quick, pick the browser account login; the flow opens quick.aws.com/sn and stores tokens in the plugin's config directory.
  • Local credentials: install and sign in to the Amazon Quick desktop app. On Windows the plugin reads the session directly from Credential Manager; on other platforms or on read failure it starts the bundled outpost. The app does not need to stay running, but you must sign in once.

Then install and log in:

magpie plugin add magpie-aws-quick
magpie login aws-quick

You can also install it from magpie's Plugins UI by searching for "Amazon Quick". After login, the account card and Usage page show plan quota, used agent hours, and trial expiry.

For clients outside magpie, node bin/serve.mjs exposes the same adapter as an OpenAI-compatible endpoint on 127.0.0.1:3426/v1 by default, and can be registered with magpie provider add --base-url http://127.0.0.1:3426/v1. Setting PROXY_TOKEN for authentication is recommended.

Usage constraints

  • Personal account only. The plugin drives the desktop client's internal WebSocket protocol rather than a public API, and the README warns this very likely violates AWS terms of service; accounts may be limited or banned.
  • OAuth is consumer/social only. Enterprise-IdP tenants may not complete the browser flow; use local credentials on a machine with the app installed instead.
  • Tool calls are simulated. Quick's own tools run server-side and are never mapped to OpenAI tool_calls; client tools are injected as a <tool_call> text convention and depend on the model following it.
  • Sampling parameters are ignored. temperature and similar fields do not exist in the protocol.
  • Images and size limits. Small inline images are sent as attachments; larger ones (up to 3 MiB each, 4 per turn, 5 MiB total) use Quick's upload flow and land in the account's cloud workspace. Remote URLs and non-image files are unsupported. The whole user_message frame is capped at 64 KB server-side, with automatic compaction of oversized prompts, schemas, and history.
  • Threads are account-bound and single-flight. Concurrent requests on the same thread are rejected; different threads can run in parallel.
  • Node ≥ 22 for tests only. The plugin itself is pure ESM with no third-party dependencies. License is MIT; the project is not affiliated with Amazon/AWS.

HotRepo's take

The repository appeared on HotRepo's new-project list for JavaScript on 2026-11-22 with 106 stars and 18 forks gained in the reporting window, but it has no recorded GitHub daily rank. That is a single-snapshot signal, so it only establishes that the project drew early attention, not that it is widely adopted. The design is unusually specific — reusing the desktop app's bundled agent and modeling its chat protocol, with a documented protocol file — which makes it more of a targeted adapter for magpie users who already have a Quick account than a general-purpose model gateway.

Written by AI from the project's public information and HotRepo's records, updated . The project's own docs are the source of truth.

You might also like

  1. aws/agent-toolkit-for-awsGives AI coding agents AWS-aware tools, skills and guardrails so they can build, deploy and manage cloud applications.Python3,166
  2. yetone/magpieRoute multiple coding agents through one local gateway, switching models, providers and accounts from a menu bar app or CLI.Go31,109