π€ ‫ π PartyKit AI
Build AI-powered real-time collaborative applications with PartyKit AI. Powered by Cloudflare AI, PartyKit brings you models for a veriety of use cases, including: text and image generation, translation, text-to-speech, and more. It also includes a vector database (powered by Cloudflare Vectorize) to build search engines and RAG applications.
Models: Text, Images, and more
To get started with using an AI model, first install the partykit-ai
package in your PartyKit project:
npm install partykit-ai
Then, import the partykit-ai
package in your server code:
import { Ai } from "partykit-ai";import type * as Party from "partykit/server";
export default class Server implements Party.Server { ai: Ai; constructor(public room: Party.Room) { this.ai = new Ai(room.context.ai); } onConnect(connection: Party.Connection) { // ... }}
You can also use the package for non-party, regular api endpoints with onFetch / onSocket / onCron:
export default class { static onFetch(request, lobby, ctx) { const ai = new Ai(lobby.ai); // ... call models }}
The partykit-ai
package is a wrapper on top of @cloudflare/ai
, so you can use the same API to access the models. For example, you could use the text-generation model to build a chatbot as described here. Learn more at the Workers AI documentation.
As an example, hereβs a fetch handler that uses the text-generation model to get a description for a word:
import { Ai } from "partykit-ai";import type * as Party from "partykit/server";
export default class { static async onFetch(request: Party.Request, lobby: Party.FetchLobby) { const ai = new Ai(lobby.ai); const result = await ai.run("@cf/meta/llama-2-7b-chat-int8", { messages: [ { role: "system", content: "You are a friendly assistant" }, { role: "user", content: "What is the origin of the phrase Hello, World" } ], stream: true });
return new Response(result, { headers: { "content-type": "text/event-stream" } }); }}
Vectorize: Build your own search engine
PartyKit AI includes a vector database (powered by Cloudflare Vectorize) to build search engines and RAG applications.
Commands
You can list all available commands by running npx partykit vectorize
in your terminal.
π PartyKit------------Usage: partykit vectorize [options] [command]
Manage vectorize indexes
Options: -h, --help display help for command
Commands: create [options] <name> Create a vectorize index delete [options] <name> Delete a vectorize index get [options] <name> Get a vectorize index by name list [options] List all vectorize indexes insert [options] [name] Insert vectors into a Vectorize index
create
Create a vectorize index
npx partykit vectorize create my-index --dimensions <dimensions> --metric <type>
# or
npx partykit vectorize create my-index --preset <preset># where <preset> is one of:# - @cf/baai/bge-small-en-v1.5# - @cf/baai/bge-base-en-v1.5# - @cf/baai/bge-large-en-v1.5# - openai/text-embedding-ada-002
delete
Delete a vectorize index
npx partykit vectorize delete my-index
get
Get a vectorize indexβ details by name
npx partykit vectorize get my-index
list
List all vectorize indexes
npx partykit vectorize list
insert
Insert vectors into a Vectorize index
npx partykit vectorize insert my-index --file <filename>
API
You can also interact with the vectorize index from your server code. After configuring your index in partykit.json
like so:
{ // ... "vectorize": { "myIndex": { "index_name": "my-index" } }}
You can access it from your server code like so:
import type * as Party from "partykit/server";
export default class implements Party.Server { constructor(public room: Party.Room) {} async onConnect(connection: Party.Connection) { const myIndex = this.room.context.vectorize.myIndex; // ... call functions on myIndex }}
// OR for non-party, regular api endpoints with onFetch / onSocket / onCron
export default class { static onFetch(request, lobby, ctx) { const myIndex = lobby.vectorize.myIndex; // ... call functions on myIndex }}
Vectors
A vector represents the vector embedding output from a machine learning model.
id
- a uniquestring
identifying the vector in the index. This should map back to the ID of the document, object or database identifier that the vector values were generated from.namespace
- an optional partition key within a index. Operations are performed per-namespace, so this can be used to create isolated segments within a larger index.values
- an array ofnumber
,Float32Array
, orFloat64Array
as the vector embedding itself. This must be a dense array, and the length of this array must match the dimensions configured on the index.metadata
- an optional set of key-value pairs that can be used to store additional metadata alongside a vector.
let vectorExample = { id: "12345", values: [32.4, 6.55, 11.2, 10.3, 87.9], metadata: { key: "value", hello: "world", url: "r2://bucket/some/object.json" }};
insert
Insert vectors into a Vectorize index
await myIndex.insert([ { id: "1", values: [1, 2, 3] }, { id: "2", values: [4, 5, 6] }]);
upsert
Upsert vectors into a Vectorize index
await myIndex.upsert([ { id: "1", values: [1, 2, 3] }, { id: "2", values: [4, 5, 6] }]);
query
Query a Vectorize index
const result = await myIndex.query( [1, 2, 3], // generate this vector with an embedding model { topK: 15, returnValues: false, returnMetadata: true });
// result is an array of { vectorId: string, score: number } objects
You can also query by namespace:
const result = await myIndex.query( [1, 2, 3], // generate this vector with an embedding model { topK: 15, returnValues: false, returnMetadata: true, namespace: "my-namespace" });
Further, you can filter results by metadata.
getByIds
Get vectors by ids
const result = await myIndex.getByIds(["1", "2"]);
deleteByIds
Delete vectors by ids
await myIndex.deleteByIds(["1", "2"]);
describe
Retrieves the configuration of a given index directly, including its configured dimensions and distance metric.
const result = await myIndex.describe();
Metadata Filtering
In addition to providing an input vector to your query, you can also filter by vector metadata associated with every vector. Query results only include vectors that match filter
criteria, meaning that filter
is applied first, and topK
results are taken from the filtered set.
By using metadata filtering to limit the scope of a query, you can filter by specific customer IDs, tenant, product category or any other metadata you associate with your vectors.
Supported operations
Optional filter
property on query()
method specifies metadata filter:
Operator | Description |
---|---|
$eq | Equals |
$ne | Not equals |
filter
must be non-empty object whose compact JSON representation must be less than 2048 bytes.filter
object keys cannot be empty, contain" | .
(dot is reserved for nesting), start with$
, or be longer than 512 characters.filter
object non-nested values can bestring
,number
,boolean
, ornull
values.
Namespace versus metadata filtering
Both namespaces and metadata filtering narrow the vector search space for a query. Consider the following when evaluating both filter types:
- A namespace filter is applied before metadata filter(s).
- A vector can only be part of a single namespace. Vector metadata can contain multiple key-value pairs. Metadata values support different types (
string
,boolean
, and others), therefore offering more flexibility.
Valid filter
examples
Implicit $eq
operator
{ "streaming_platform": "netflix" }
Explicit operator
{ "someKey": { "$ne": true } }
Implicit logical AND
with multiple keys
{ "pandas.nice": 42, "someKey": { "$ne": true } }
Keys define nesting with .
(dot)
{ "pandas.nice": 42 }
// looks for { "pandas": { "nice": 42 } }
Examples
Add metadata
With the following index definition:
$ npx partykit vectorize create tutorial-index --dimensions=3 --metric=cosine
Metadata can be added when inserting or upserting vectors.
const newMetadataVectors = [ { id: "1", values: [32.4, 74.1, 3.2], metadata: { url: "/products/sku/13913913", streaming_platform: "netflix" } }, { id: "2", values: [15.1, 19.2, 15.8], metadata: { url: "/products/sku/10148191", streaming_platform: "hbo" } }, { id: "3", values: [0.16, 1.2, 3.8], metadata: { url: "/products/sku/97913813", streaming_platform: "amazon" } }, { id: "4", values: [75.1, 67.1, 29.9], metadata: { url: "/products/sku/418313", streaming_platform: "netflix" } }, { id: "5", values: [58.8, 6.7, 3.4], metadata: { url: "/products/sku/55519183", streaming_platform: "hbo" } }];
// Upsert vectors with added metadata, returning a count of the vectors upserted and their vector IDslet upserted = await env.YOUR_INDEX.upsert(newMetadataVectors);
Query examples
Use the query()
method:
let queryVector = [54.8, 5.5, 3.1];// Best match is vector id = 5 (score closet to 1)let originalMatches = await YOUR_INDEX.query(queryVector, { topK: 3, returnValues: true, returnMetadata: true});
Results without metadata filtering:
{ "matches": [ { "id": "5", "score": 0.999909486, "values": [58.79999923706055, 6.699999809265137, 3.4000000953674316], "metadata": { "url": "/products/sku/55519183", "streaming_platform": "hbo" } }, { "id": "4", "score": 0.789848214, "values": [75.0999984741211, 67.0999984741211, 29.899999618530273], "metadata": { "url": "/products/sku/418313", "streaming_platform": "netflix" } }, { "id": "2", "score": 0.611976262, "values": [15.100000381469727, 19.200000762939453, 15.800000190734863], "metadata": { "url": "/products/sku/10148191", "streaming_platform": "hbo" } } ]}
The same query()
method with a filter
property supports metadata filtering.
let queryVector = [54.8, 5.5, 3.1];// Best match is vector id = 4 with metadata filterlet metadataMatches = await YOUR_INDEX.query(queryVector, { topK: 3, filter: { streaming_platform: "netflix" }, returnValues: true, returnMetadata: true});
Results with metadata filtering:
{ "matches": [ { "id": "4", "score": 0.789848214, "values": [75.0999984741211, 67.0999984741211, 29.899999618530273], "metadata": { "url": "/products/sku/418313", "streaming_platform": "netflix" } }, { "id": "1", "score": 0.491185264, "values": [32.400001525878906, 74.0999984741211, 3.200000047683716], "metadata": { "url": "/products/sku/13913913", "streaming_platform": "netflix" } } ]}