Last reviewed
May 9, 2026
Sources
No citations yet
Review status
Needs citations
Revision
v2 ยท 2,496 words
Add missing citations, update stale details, or suggest a clearer explanation.
See also: ChatGPT Plugins, ChatGPT Plugin Categories and Beverage
Beverage ChatGPT Plugins were the small subset of third-party plugins published in the ChatGPT plugins store between March 2023 and April 2024 that focused on drinks: cocktails, coffee, sake, and other consumables outside the savoury food remit covered by Food ChatGPT Plugins. The category was never large. At its peak in late 2023 the entire beverage corner of the plugin catalogue could be counted on one hand, with three plugins recurring in independent surveys: Drink Maestro for cocktail recipes, CoffeeHo for coffee questions, and Sakenowa for Japanese sake. The hosting program was deprecated on March 19, 2024 and shut down for end users on April 9, 2024, succeeded by Custom GPTs and the GPT Store. This article treats the category as a historical reference for an early experiment in connecting a large language model to live drinks data.
OpenAI announced ChatGPT plugins on March 23, 2023. The launch slate listed eleven external partners and two first-party plugins (a code interpreter and a browsing tool). None of the external partners was a dedicated beverage service. The closest adjacencies were Instacart (grocery cart) and Wolfram (nutritional reference data). Beverage-specific plugins arrived later, after general availability opened the store to third-party submissions.
On May 12, 2023 OpenAI announced that web browsing and plugins would become available to all ChatGPT Plus subscribers the following week. The plugin store opened for general access on May 19, 2023. The first beverage-focused entries appeared in this window. Sakenowa was approved on May 29, 2023 according to its developer. Drink Maestro surfaced in independent plugin lists by early June 2023. CoffeeHo appeared in catalogue snapshots from late July 2023.
The deprecation arrived in two steps. At OpenAI DevDay on November 6, 2023 the company introduced Custom GPTs and announced that plugins would be retired in favour of GPTs and their tool-calling layer, Actions. On March 19, 2024 OpenAI sent the formal deprecation notice, and on April 9, 2024 plugins were turned off. The official notice, "Winding down the ChatGPT plugins beta," pointed users to GPTs and the GPT Store (opened January 10, 2024) for equivalent functionality.
Key dates for the beverage category:
| Date | Event |
|---|---|
| March 23, 2023 | OpenAI announces ChatGPT plugins. The eleven external launch partners include no dedicated beverage service. |
| May 12, 2023 | OpenAI announces general availability of plugins for ChatGPT Plus subscribers. |
| May 19, 2023 | Plugin store opens to all Plus subscribers. Long-tail beverage plugins begin to appear. |
| May 29, 2023 | Sakenowa, the Japanese sake plugin, is approved and listed in the store. |
| Early June 2023 | Drink Maestro appears in independent plugin survey lists. |
| Late July 2023 | CoffeeHo appears in catalogue snapshots as a coffee expert plugin. |
| November 6, 2023 | OpenAI DevDay introduces Custom GPTs and Actions. Plugins are publicly described as a sunsetting feature. |
| January 10, 2024 | GPT Store opens, beginning the migration of plugin functionality into branded GPTs. |
| March 19, 2024 | OpenAI formally deprecates the plugins beta. |
| April 9, 2024 | The plugin store and plugin runtime are turned off for end users. |
A beverage plugin was a small piece of server software exposing an HTTP API to the chatbot. Like every plugin, it consisted of two artefacts on the developer's domain: a manifest file named ai-plugin.json served at /.well-known/ai-plugin.json, and an OpenAPI specification describing the available endpoints. ChatGPT used the manifest to describe the plugin to the GPT-4 model in natural language and used the OpenAPI document to plan and validate calls. Prompt engineering on the user side shaped which plugin the model decided to call.
The practical capabilities clustered into a small number of recurring patterns: recipe lookup (search a cocktail corpus by name, ingredient, or alphabet), brand or product search (look up a specific sake brewery or coffee bean variety), discovery and recommendation (random or trending suggestions), educational explanation (brewing methods, flavour profiles, history), and filtering by attribute (non-alcoholic, regional, flavour-based).
Almost no beverage plugin completed a transaction inside ChatGPT itself. Where the food category had Instacart-style grocery checkout flows, the beverage category was overwhelmingly informational. None of the three core beverage plugins took payment or shipped product.
Three plugins recur across independent catalogue snapshots from June 2023 through early 2024. They are the only entries in the beverage category that can be reliably cross-referenced to more than one source, so they are the only ones documented by name in this article.
Drink Maestro was a cocktail recipe plugin that styled itself "your personal bartender." Its public description in the plugin store read: "Learn to mix any drink you can imagine (real or made-up), and discover new ones. Includes drink images." The plugin was developed by Nikki Mitchell and served from drinkmaestro.space, with a public manifest at /.well-known/ai-plugin.json and an OpenAPI specification that exposed five endpoints. The endpoints were:
| Endpoint | Method | Purpose |
|---|---|---|
searchDrink | GET | Look up a cocktail by name and return its recipe and instructions. |
listCocktails | GET | Return cocktails whose names begin with a given letter. |
searchIngredient | GET | Return ingredient details and the cocktails that use a given ingredient. |
randomCocktail | GET | Return a randomly selected cocktail. |
nonAlcoholicList | GET | Return non-alcoholic drink options. |
In practice, a user might describe what they had on hand ("I have gin, lime juice, and tonic") and ChatGPT would call searchIngredient, or ask for a specific drink ("how do I make a paloma?") and ChatGPT would call searchDrink. The plugin returned drink images, which made it more visually engaging than text-only beverage tools. Drink Maestro is listed in independent plugin surveys, including the June 1, 2023 catalogue snapshot maintained by Booster Mini Class and Orren Prunckun's July 9, 2023 list.
Sakenowa was a Japanese sake search plugin built on the data of Sakenowa.com, a sake-focused community platform run from Tokyo. The plugin was developed by Akira Ueda, a Tokyo-based freelancer, who published a detailed write-up of the build process on the dev.to community blog. The plugin was approved on May 29, 2023, seven days after submission, which makes it one of the earliest beverage-specific entries in the catalogue.
The plugin's public description read: "Find Sake and get detailed information in various ways." Its OpenAPI specification exposed five endpoints, summarised below.
| Endpoint | Purpose |
|---|---|
searchBrand | Look up a sake brand by name in Japanese kanji, hiragana, or English. |
getPopularBrands | Return currently trending sake brands. |
getFlavorTags | Return the set of flavour tags Sakenowa applies to brands. |
searchByArea | Return brands from a given Japanese prefecture, addressed by official prefecture number. |
getFlavorChart | Return a six-axis radar chart of a brand's flavour profile. |
The radar-chart endpoint was distinctive. Sakenowa.com had built a six-axis flavour vector for each sake with axes labelled in Japanese (haiyaka for elegance at the top, hojun for richness next clockwise, and four more around the circle). The plugin surfaced this vector to ChatGPT, which could then describe a sake's profile in natural language without inventing the data. The plugin used a service-level Bearer token (the service_http authentication scheme), and its developer documented a deliberate design choice: the plugin would refuse to fabricate results and would direct users back to Sakenowa.com when a query returned nothing. Akira Ueda's dev.to write-up is one of the few first-hand developer accounts of the plugin approval process for a beverage plugin and records that modifying a manifest after approval triggered a fresh three-to-seven-day review.
CoffeeHo described itself as "your coffee expert," or in some directory listings "the ultimate coffee guru: your personal espresso connoisseur." The plugin appeared in catalogue snapshots from late July 2023 and was documented as a working plugin through to the end of the program in April 2024. CoffeeHo was a question-answering plugin rather than a search plugin: it accepted free-form coffee questions and returned guidance grounded in coffee-specific reference material. Topics it was cited as covering included brewing methods (espresso, pour-over, French press, AeroPress, cold brew), bean varieties and origins, the history of coffee, and equipment recommendations. First-party documentation is sparser than for Drink Maestro or Sakenowa, so the article does not enumerate its API endpoints by name.
Several entries are sometimes informally described as beverage plugins but do not actually fit the category.
| Item | Why it does not count |
|---|---|
| Mocktail Mixologist | A Custom GPT in the post-November 2023 GPT Store, not a plugin. |
| BarGPT | Standalone cocktail web app built on the OpenAI API. Never published as a plugin. |
| Vivino, Wine.com | No mainstream wine retailer or rating service published a plugin during the plugin era. |
| OpenTable, Tabelog, Hot Pepper Gourmet | Restaurant plugins categorised under Food ChatGPT Plugins. |
| Klook | Travel and activities plugin, not a beverage plugin. |
Across the eleven months when the plugin store was actively curated, the verifiable beverage-only plugins were Drink Maestro, Sakenowa, and CoffeeHo. The narrowness was characteristic of niches that could not lean on a household-name retail brand.
Two workflow patterns dominated the beverage category. The cocktail recipe lookup flow used Drink Maestro: the user described a drink by name or by available ingredients, ChatGPT routed the query to searchDrink, searchIngredient, or randomCocktail, and the plugin returned a structured recipe with an image. The regional drinks search flow used Sakenowa: the user asked for a sake by brand, region, or flavour, ChatGPT routed the query to searchByArea or searchBrand plus getFlavorChart, and Sakenowa returned a structured response including the six-axis flavour vector for the model to translate into natural-language tasting notes.
A recurring complaint with both flows was silent fallback. ChatGPT often answered drink questions from training data instead of calling the plugin, particularly for common cocktails the base model already knew. From the user's view, the chat answer looked the same whether it came from the plugin or from the model's own knowledge, which made it hard to tell when the live data path was working.
OpenAI gave several public reasons for ending the plugin program. The beverage category illustrated most of them in compressed form because it was so small to begin with. Discovery was poor: with a flat catalogue and only a handful of beverage entries, users still struggled to find them, and Drink Maestro, Sakenowa, and CoffeeHo did not appear together in any single curated section. The three-plugin-per-conversation limit constrained workflows that wanted a recipe plugin plus a food plugin for pairing plus a nutrition plugin. Reliability suffered because the model often answered drink questions from training data instead of calling the plugin, especially for common cocktails. The single-purpose limit meant a plugin could only expose API endpoints with no persistent instructions or identity, so a user looking for a "home bartender assistant" had to assemble one from raw API calls. Finally, the plugin program offered no revenue share, while the successor GPT Store launched with revenue-sharing intentions that pulled developers across.
The successor to the plugin program is the combination of Custom GPTs and the GPT Store. A GPT bundles three things: custom instructions, optional knowledge files, and Actions (the new name for the plugin-style API call layer). Where the plugin store presented three or four single-purpose beverage plugins, the GPT Store of 2024 presented many more drinks-themed assistants under names like Cocktail GPT, Mocktail Mixologist, Drinks, Wine, Coffee, and CoffeeHo Coffee GPT. Most used the same HTTP backend pattern (or no backend at all, relying on the base model plus instructions). Discovery moved from a flat list to a marketplace of named applications, which suited the niche better: a curious cocktail enthusiast could find a bartender GPT immediately rather than having to know Drink Maestro by name.
The Actions layer underneath GPTs uses the same OpenAPI specification format that plugins used, so the few beverage plugin developers who continued operating in 2024 had a low-cost migration path: keep the OpenAPI spec, repackage behind a Custom GPT, ship to the store. The table below contrasts the beverage category before and after the transition.
| Aspect | Plugin era (2023 to 2024) | GPT era (2024 onward) |
|---|---|---|
| Beverage entries | Drink Maestro, Sakenowa, CoffeeHo | Drinks GPTs, cocktail GPTs, mocktail GPTs, wine GPTs, coffee GPTs |
| Discovery | Flat plugin store inside ChatGPT | Branded GPTs in the GPT Store |
| Activation | User toggled up to three plugins per chat | User opened a single GPT |
| Backend contract | OpenAPI plus ai-plugin.json manifest | OpenAPI plus a GPT configuration |
| Persistence of identity | None; plugins were stateless API wrappers | GPTs carry instructions and knowledge across sessions |
| Revenue model | None for developers | Revenue share announced for GPT Store |
manifests/drinkmaestro.space.json and specs/drinkmaestro.space.yaml.drinkmaestro.space.