@lore-co/bedrock

Host-side Lore integration for applications deployed on Amazon Bedrock AgentCore. Bedrock is a model and agent platform, not a local coding-agent executable, so this adapter runs at the application's invocation boundary.

import { BedrockAgentCoreAdapter } from "@lore-co/bedrock";

const lore = new BedrockAgentCoreAdapter({
  baseUrl: process.env.LORE_API_URL!,
  headers: {
    authorization: `Bearer ${process.env.LORE_WORKSPACE_TOKEN!}`,
  },
});

const prepared = await lore.prepareInvocationReliable({
  eventId: request.id,
  sessionId: request.sessionId,
  prompt: request.prompt,
  previousAssistant: request.previousAssistant,
  repo: request.repository,
});

const result = await agent(prepared.prompt);

Persist the last assistant response in the host's session state and pass it as previousAssistant on the next invocation. The adapter records the paired turn, applies Lore's redaction policy, retrieves scoped knowledge, records a delivery receipt, and returns the enriched prompt. It works with Bedrock model providers and AgentCore-hosted frameworks because it does not depend on a specific model SDK. prepareInvocationReliable uses Reliability v1 capture and delivery responses; prepareInvocation remains available for legacy servers. The package exports BEDROCK_RELIABILITY_INTEGRATION for machine-readable capability discovery.