Adding Amazon Bedrock as an embedding provider

[doyeonyeah]

Summary

The plugin supports amazon-bedrock for chat models but not for embedding models. Could Bedrock be added as an embedding provider, covering all embedding models available on Bedrock (Cohere, Amazon Titan, and any future additions)?

Motivation

My organization only has model access through Amazon Bedrock — direct Cohere / OpenAI keys aren't available to me. Bedrock hosts several embedding model families that would be useful for Copilot's vault indexing:

• Cohere — cohere.embed-multilingual-v3, cohere.embed-english-v3, and embed-v4.0 in supported regions
• Amazon Titan — amazon.titan-embed-text-v2:0, amazon.titan-embed-text-v1, amazon.titan-embed-image-v1
• Any future embedding models AWS adds (the same invoke pattern works for all of them)

Today the embedding provider dropdown offers OpenAI, Azure OpenAI, Cohere AI (direct), Google, Ollama, Mistral, Copilot Plus, 3rd-party OpenAI-format, and Jina — but no Bedrock. The chat side already has all the Bedrock plumbing (BedrockChatModel, SigV4, region handling, API-key auth), so embeddings should be
able to follow the same pattern.

Proposal

• Add amazon-bedrock to the embedding provider list.
• Reuse the existing Bedrock auth/region settings so users don't have to re-enter credentials.
• Let users enter any Bedrock embedding model ID (not a hardcoded dropdown), since Bedrock's catalog expands frequently. A few suggested IDs in the placeholder would be enough:
  • cohere.embed-multilingual-v3
  • amazon.titan-embed-text-v2:0
• Dispatch request/response shape by provider prefix (cohere.* vs amazon.*) — Titan and Cohere use slightly different payloads, but both are straightforward InvokeModel calls.

Workarounds considered

• Direct Cohere / OpenAI — not available to me (enterprise-only Bedrock access).
• Local OpenAI-format proxy in front of Bedrock — works but adds a process to maintain.

Happy to help test. Thanks for the plugin!

One small technical note worth keeping in mind (not for the issue body, just for you): Cohere and Titan use different request JSON on Bedrock — Cohere wants {texts: [...], input_type: "search_document"}, Titan wants {inputText: "..."} one at a time. Whoever implements it will need to branch on the model ID
prefix. That's why the "any model ID" ask in the proposal matters.