Connect Positron to AWS Bedrock - Anthropic

[axhxu]

Hi,
We are looking to test positron in our authoring environments but want to use AWS Bedrock AI platform with Anthropic Claude.
I found this post from last year post74640.html?hilit=AI%20Bedrock#p74640
Just wondering if there are more details on how to do this?
We talked to AWS and they suggested the following: Positron → OpenAI connector → AWS Bedrock Access Gateway → Claude on AWS Bedrock
The AWS Bedrock Access Gateway would have the OpenAI api and connect to Claude.
Would I still use the MS Azure OpenAI connector of Positron Assistant Enterprise as the OpenAI connector?
Are there details on adding a authorization header?

[Cosmin Duna]

Hi
The AWS Bedrock Access Gateway was not available last year, so it's great you're exploring it now. According to the specifications on GitHub (https://github.com/aws-samples/bedrock-access-gateway), the gateway requires a Bearer authorization header. I recommend using the OpenAI connector, as it automatically includes the key like this in the request. The MS Azure connector uses the api-key HTTP header, which is different.

Here's what you should do:
1 . Add the API Base URL into the Address field. It should look like this: http://xxxx.xxx.elb.amazonaws.com/api/v1.
2. Enter the key in the "API key" field.
3. Disable text moderation by unchecking the "Enable text moderation" checkbox, as the gateway doesn't support this request type.

Please let me know if this setup works for you.

Best regards,
Cosmin

[liyan]

Hi Cosmin
Sorry to jump in to this thread.
I have the same issue as axhxu.
We have an internal deepseek model using beartoken. As you said, using openai connector can connect the AI Positron to other model or gateway.
But what you suggested to do still requires a key.
As you said in step 2, enter a key in "API key" field.
We are using beartoken, how do we have such key?

Regards
liyan

[Radu]

Adding a link to the separate discussion with Lyan about bearer tokens:

post76607.html#p76607

[axhxu]

The Bedrock Gateway and using the OpenAI connector worked like a charm.
Just have a question if there is a preferred Claude Model?
I selected Sonnet 3.7 but wondering if one of the older models would be better?
Do I need a model with Text and Vision or just text?

[Radu]

Hi,
I think Claude Sonnet 3.7 should be perfectly fine, ideally you should start using our builtin actions on small sample documents and see how they behave. We have a sample project with small topics on which various AI actions can be applied in order to see how they work:
https://github.com/oxygenxml-incubator/ ... nt-samples

And yes, support for Vision would be important, we have several actions like, "Generate Documentation Draft", "Update Content based on images" or "Generate Image Alternate Text" which require vision support.

Regards,
Radu