import {
createAction,
Property,
StoreScope,
} from '@activepieces/pieces-framework';
import {
ConverseCommand,
ConversationRole,
Message,
} from '@aws-sdk/client-bedrock-runtime';
import { awsBedrockAuth } from '../../index';
import {
createBedrockRuntimeClient,
getBedrockModelOptions,
formatBedrockError,
extractConverseTextResponse,
} from '../common';
export const sendPrompt = createAction({
auth: awsBedrockAuth,
name: 'send_prompt',
displayName: 'Ask Bedrock',
description: 'Send a text prompt to an Amazon Bedrock model.',
props: {
model: Property.Dropdown({
displayName: 'Model',
required: true,
auth: awsBedrockAuth,
description: 'The foundation model to use for generation.',
refreshers: [],
options: async ({ auth }) => {
if (!auth) {
return {
disabled: true,
placeholder: 'Connect your AWS account first',
options: [],
};
}
return getBedrockModelOptions(auth.props, { useInferenceProfiles: true });
},
}),
prompt: Property.LongText({
displayName: 'Prompt',
required: true,
}),
systemPrompt: Property.LongText({
displayName: 'System Prompt',
required: false,
description: 'Instructions that guide the model behavior.',
}),
temperature: Property.Number({
displayName: 'Temperature',
required: false,
description:
'Controls randomness. Lower values produce more deterministic output.',
defaultValue: 0.7,
}),
maxTokens: Property.Number({
displayName: 'Maximum Tokens',
required: false,
description: 'The maximum number of tokens to generate.',
defaultValue: 2048,
}),
topP: Property.Number({
displayName: 'Top P',
required: false,
description:
'Nucleus sampling: the model considers tokens with top_p probability mass.',
defaultValue: 1,
}),
stopSequences: Property.Array({
displayName: 'Stop Sequences',
required: false,
description:
'Sequences that will cause the model to stop generating. Up to 4 sequences.',
}),
memoryKey: Property.ShortText({
displayName: 'Memory Key',
required: false,
description:
'A memory key that will keep the chat history shared across runs and flows. Keep it empty to leave the model without memory of previous messages.',
}),
},
async run({ auth, propsValue, store }) {
const client = createBedrockRuntimeClient(auth.props);
const {
model,
prompt,
systemPrompt,
temperature,
maxTokens,
topP,
stopSequences,
memoryKey,
} = propsValue;
let messageHistory: Message[] = [];
if (memoryKey) {
messageHistory =
(await store.get<Message[]>(memoryKey, StoreScope.PROJECT)) ?? [];
}
messageHistory.push({
role: ConversationRole.USER,
content: [{ text: prompt }],
});
try {
const response = await client.send(
new ConverseCommand({
modelId: model,
messages: messageHistory,
...(systemPrompt ? { system: [{ text: systemPrompt }] } : {}),
inferenceConfig: {
temperature: temperature ?? undefined,
maxTokens: maxTokens ?? undefined,
topP: topP ?? undefined,
stopSequences:
stopSequences && stopSequences.length > 0
? (stopSequences as string[])
: undefined,
},
})
);
const outputMessage = response.output?.message;
if (outputMessage) {
messageHistory.push(outputMessage);
}
if (memoryKey) {
await store.put(memoryKey, messageHistory, StoreScope.PROJECT);
}
return extractConverseTextResponse(response);
} catch (error) {
throw new Error(formatBedrockError(error));
}
},
});
