Support Anthropic Messages API in the Custom post-processing provider

[diligentpig]

Hi @cjpais ,
Thanks for your continued work on Handy!

I use Claude Haiku 5.5 hosted on Microsoft Foundry and would like to use it for post-processing in Handy. Foundry exposes an Anthropic Messages API endpoint, while Handy’s Custom provider currently uses OpenAI-compatible Chat Completions.

• This is related to #1386. I saw the comment explaining that this is a feature request rather than a bug, and that a PR could be submitted. I wanted to check whether this is still something you would consider before working on it.
• I also saw the request in #168 from someone whose workplace only allows Amazon Bedrock. Their proposed solution was different, but it is another example of needing to use a specific hosting service for post-processing.

My suggestion is to add an new "API format" option to the Custom provider:

• OpenAI-compatible (Chat Completions) — the default, keeping the current behavior.
• Anthropic Messages — an optional format for compatible endpoints.

Users would keep using the existing Base URL, API key, and model fields. The new format would handle the Messages endpoint, headers, request body, and response parsing. I would keep the scope to basic text post-processing with API key authentication.

My main use case is Microsoft Foundry, but this could help with other services using the same format. For example, Amazon Bedrock also documents Anthropic-compatible Messages endpoints.

• I think the Custom provider is a useful place for this because users could choose the API format without needing a separate provider entry for each hosting service. If another format is needed later, it could use the same setting.
• Currently, I'm using LiteLLM as a proxy to translate between the two formats, but a direct connection would mean one less service to set up.

I also read #1581 about moving Handy toward local-only post-processing. The "API format" option can apply to local servers too—for example, Ollama supports a local Anthropic-compatible Messages endpoint. Still, I understand that built-in local inference may not need an external API at all, so this may not be useful for general users in the long term (but could be an option for advanced users?).

Given that direction and the feature freeze, would you consider a focused PR for this? I’m happy to help with implementation and testing on my Foundry deployment if the scope makes sense.

[cjpais]

If you make a PR for it, I'll at least review it, but just note I may close it. I haven't made a decision on this concretely yet, but I don't mind tracking it in a PR for right now, and then I can make a decision on it then. I assume it probably as long as it's like simple and keeps new UI additions minimal. I don't really see a big problem

The only thing is to note, though, is that if there are any bugs, I will be tagging you in that thread to help resolve them

[diligentpig]

Thanks @cjpais! I've opened #2257 for this.

I kept it to a single "API Format" dropdown that only appears for the Custom provider, with OpenAI Compatible as the default, so existing setups and the current Chat Completions path are unchanged. Happy to adjust anything, and I'll help out if any bugs come up.