Introducing Grafbase KV

Today we're excited to announce the release of Grafbase KV!

Grafbase KV offers a straightforward and speedy serverless key-value storage solution for resolvers. It boasts eventual consistency, edge deployment, and eliminates the need for configuration both locally and in production. The methods it currently supports are:

• list (with prefix filtering and cursor-based pagination)
• get
• set (with metadata/time-to-live options)
• delete

Grafbase KV is experimental, and you can enable it for your project using the SDK:

import { config, graph } from '@grafbase/sdk'

const g = graph.Standalone()

export default config({
  graph: g,
  experimental: {
    kv: true,
  },
})

Once enabled, you can invoke kv from the third context argument inside Edge Resolvers:

export default async function MyResolver(_, { prompt }, { kv }) {
  // ...
}

Here's an example of storing the response for a prompt sent to OpenAI:

import { OpenAI } from 'langchain/llms/openai'

export default async function HelloResolver(_, { prompt }, { kv }) {
  const model = new OpenAI({
    modelName: 'gpt-3.5-turbo-instruct',
    openAIApiKey: process.env.OPENAI_API_KEY,
  })

  try {
    const { value } = await kv.get(prompt)

    if (value === null) {
      const response = await model.invoke(prompt)

      await kv.set(prompt, response, { ttl: 60 })

      return response
    } else {
      return value
    }
  } catch (e) {
    return 'error'
  }
}

When working locally we can already see the advantages of using Grafbase KV. We are able to store the prompt and instruct response as a key-value pair that significantly improves response times globally, as well as saving costs with OpenAI by not running the same prompt repeatedly.

You can learn more about how all of this works by exploring the documentation.

We'd love to hear your feedback and ideas, so join please us on Discord.