[BUG] Support for AWS Bedrock Application Inference Profiles

[botoxparty]

Issue

Roo Code doesn't support AWS Bedrock Application Inference Profiles, which work fine in Claude Code and AWS CLI with the same credentials.

Error Message

Error: You don't have access to the model specified.

What Works

AWS CLI:

aws bedrock-runtime invoke-model \
  --model-id arn:aws:bedrock:eu-west-1:xxxxxxxxxxxx:application-inference-profile/xxxxxxxxxxxxx \
  --region eu-west-1 \
  --profile my-profile

✅ Works perfectly

Claude Code:
✅ Works with the same application inference profile

What Doesn't Work

Roo Code:

{
  "aws.profile": "my-profile",
  "aws.region": "eu-west-1",
  "bedrock.modelId": "arn:aws:bedrock:eu-west-1:xxxxxxxxxxxx:application-inference-profile/xxxxxxxxxxxxx"
}

❌ Fails with access error

Why This Matters

Many enterprise AWS accounts restrict access to foundation models and only allow application inference profiles for centralized control. Roo Code should support these like Claude Code does.

Request

Please add support for application inference profile ARNs in the format:
arn:aws:bedrock:region:account:application-inference-profile/profile-id

Context (who is affected and when)

Happens when i try to chat with roocode after configuring my settings.

Reproduction steps

1. MacOS, latest extension version
2. Configure AWS Profile
3. Settings were applied but inference fails

Expected result

RooCode can use the application inference profile

Actual result

Error: You don't have access to the model specified.

Variations tried (optional)

No response

App Version

3.50.5

API Provider (optional)

Amazon Bedrock

Model Used (optional)

application inference profile

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

[roomote]

I've analyzed this issue and here's my proposed implementation plan:

Root Cause Analysis

After tracing through the code, I found that the ARN parsing regex in src/api/providers/bedrock.ts already matches application-inference-profile ARNs correctly, and the custom ARN code path properly passes the full ARN to the Bedrock API. The likely problem is a combination of discoverability and documentation issues -- the "Use Custom ARN..." option in the model selector is required for ARN-based model IDs, but the UI and docs don't mention application-inference-profile as a supported type.

There's also a potential UX issue: if a user enters an ARN directly in the model ID field (instead of the custom ARN field), the code silently falls back to the default model, which would cause the "access denied" error if the account only has access via application inference profiles.

Proposed Changes

1. Add application-inference-profile to the UI examples (webview-ui/src/components/settings/providers/BedrockCustomArn.tsx): Add an example application-inference-profile ARN to the list of format examples shown under the custom ARN field, so users know this is a supported format.

2. Add application-inference-profile to parseArn documentation (src/api/providers/bedrock.ts): Update the code comments listing supported ARN types to include Application Inference Profile.

3. Improve model info heuristics for application inference profiles: In guessModelInfoFromId, application inference profile IDs may be UUIDs or custom names that don't match any model pattern. The code should handle application-inference-profile model types by providing reasonable defaults (e.g., enabling supportsImages and supportsPromptCache since the underlying model is likely Claude).

4. Add test cases: Add unit tests for application-inference-profile ARNs in the existing test files (bedrock-custom-arn.spec.ts and bedrock.spec.ts).

5. Update error message for ACCESS_DENIED: Mention application inference profiles in the error guidance to help users who encounter this error find the correct configuration option.

This approach ensures backward compatibility while making application inference profiles a first-class supported feature with clear guidance.

Please mention @roomote in a comment saying that you approve this plan, and I'll proceed with the implementation.

[botoxparty]

@roomote i approve this plan

[roomote]

I've seen the approval and will start implementing the plan now. I'll open a draft PR to share progress shortly.

[roomote]

Opened a draft PR with my attempt at addressing this: #11766

The changes add first-class support for application-inference-profile ARNs including UI examples, improved model defaults, better error guidance, and test coverage. Feedback welcome!