feat: Add AWS Bedrock Inference Profile ARN resolution support

[marcelloceschia]

This PR adds full support for AWS Bedrock inference profiles (both application-inference-profile and inference-profile ARNs) with automatic model resolution to detect underlying model capabilities. This fixes an issue where prompt caching was not being enabled for inference profile ARNs because the extension couldn't
determine the underlying model's capabilities.

Implementation

The implementation includes:

1. New Bedrock Inference Profile Resolver (src/api/providers/bedrock-inference-profile-resolver.ts)

   • Added @aws-sdk/client-bedrock dependency for AWS Bedrock API access
   • Resolves application inference profile ARNs to underlying model ARNs
   • Resolves standard inference profile ARNs to model IDs
   • Caches resolved models to minimize API calls
   • Handles AWS credential configuration automatically

2. Enhanced Bedrock Provider (src/api/providers/bedrock.ts)

   • Integrated inference profile resolver into the Bedrock provider
   • Automatic detection of model capabilities (prompt caching, extended context, reasoning budgets) from resolved models
   • Backward compatible - works with direct model IDs and ARNs

3. Improved Settings UI

   • Added real-time ARN resolution feedback in the Bedrock custom ARN settings (webview-ui/src/components/settings/providers/BedrockCustomArn.tsx)
   • Shows resolved model information when entering inference profile ARNs
   • Enhanced user experience with visual feedback during resolution

4. Comprehensive Test Coverage (src/api/providers/__tests__/)

   • Unit tests for both application inference profiles and standard inference profiles
   • Tests for caching behavior and error handling
   • Tests for capability detection from resolved models

Screenshots

before	after
Inference profile ARNs were treated as unknown models, prompt caching disabled	Inference profiles are resolved to underlying models with full capability detection

How to Test

1. Configure AWS Bedrock provider in settings
2. Use the "Custom ARN" option and enter an inference profile ARN, such as:
   • Application inference profile: arn:aws:bedrock:us-east-1:123456789012:application-inference-profile/my-profile
   • Standard inference profile: arn:aws:bedrock:us-west-2:inference-profile/us.anthropic.claude-3-5-sonnet-20241022-v2:0
3. Verify that the UI shows the resolved model information
4. Start a conversation and confirm that prompt caching works correctly (check API usage logs)
5. Verify that model capabilities (extended context, reasoning budgets) are correctly detected

[changeset-bot]

🦋 Changeset detected

Latest commit: 8f26ece

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 12 packages

Name	Type
kilo-code	Minor
@roo-code/types	Patch
@kilocode/cli	Patch
@roo-code/web-evals	Patch
@roo-code/web-roo-code	Patch
@kilocode/agent-runtime	Patch
@roo-code/cloud	Patch
@kilocode/core-schemas	Patch
@roo-code/core	Patch
@roo-code/evals	Patch
@roo-code/ipc	Patch
@roo-code/telemetry	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[Githubguy132010]

Not needed perse, but would prefer to have a more updated model as default instead of sonnet 3.5.

[marcelloceschia]

I will update the list after this is merged. Our aws solution architect gave us also an hint how the 1M context window works for sonnet 4.5 and opus 4.5

[Githubguy132010]

I will update the list after this is merged. Our aws solution architect gave us also an hint how the 1M context window works for sonnet 4.5 and opus 4.5

Alright LGTM then.

[marcelloceschia]

Thank you @kevinvandijk for taking your time to review this changes. Feel free to suggest further changes