feat: add AWS Bedrock LLM provider via native Converse API

[cgorski]

No description provided.

[gemini-code-assist]

Summary of Changes

Hello @cgorski, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request introduces a significant new feature by integrating AWS Bedrock as a first-class LLM provider. This allows users to leverage Bedrock's capabilities, including its native Converse API and diverse model offerings, within the application. The implementation focuses on providing a robust and flexible authentication experience, accommodating various AWS credential management strategies, and ensuring a smooth setup process through updated configuration and wizard flows.

Highlights

• New LLM Provider: AWS Bedrock: Added support for AWS Bedrock as a new Large Language Model (LLM) provider, leveraging its native Converse API for interactions.
• Flexible Authentication for Bedrock: Implemented support for various AWS Bedrock authentication methods, including bearer tokens, IAM credentials, SSO profiles, and EC2/ECS instance roles, all handled transparently by the AWS SDK credential chain.
• Native Converse API Integration: Integrated directly with the AWS Bedrock Converse API, allowing for advanced features like system messages, tool use, and robust error handling specific to Bedrock.
• Enhanced Configuration and Setup: Updated documentation, configuration structures, and the interactive setup wizard to seamlessly guide users through setting up AWS Bedrock, including region selection, cross-region inference, and model ID specification.

Changelog

• CHANGELOG.md
  • Added an entry for the new AWS Bedrock LLM provider with native Converse API support.
• CLAUDE.md
  • Updated documentation to include configuration examples for AWS Bedrock, detailing environment variables for region, model ID, cross-region settings, and authentication methods.
• Cargo.lock
  • Updated dependency lock file to include numerous new AWS SDK crates and their transitive dependencies required for AWS Bedrock integration.
• Cargo.toml
  • Added new dependencies for AWS Bedrock integration, specifically aws-config, aws-sdk-bedrockruntime, and aws-smithy-types.
• FEATURE_PARITY.md
  • Updated the feature parity table to reflect full support (✅) for AWS Bedrock as an LLM provider and for tool usage.
• docs/LLM_PROVIDERS.md
  • Added AWS Bedrock to the list of supported LLM providers, including its backend identifier and authentication methods.
  • Included a new detailed section for AWS Bedrock, explaining how to configure it with API keys, AWS credentials, and cross-region inference, along with popular model IDs.
• src/config/llm.rs
  • Added a Bedrock variant to the LlmBackend enum to represent the new provider.
  • Introduced BedrockConfig struct to hold specific configuration parameters for AWS Bedrock, such as region, model, cross-region settings, and AWS profile.
  • Modified the LlmConfig::from_env_and_settings method to parse and populate BedrockConfig when LLM_BACKEND is set to bedrock.
• src/config/mod.rs
  • Exported the new BedrockConfig struct.
  • Added bedrock_api_key to the list of secrets that can be injected from the secrets store.
• src/llm/bedrock.rs
  • Added a new module implementing BedrockProvider for AWS Bedrock's native Converse API.
  • Implemented message conversion logic to adapt internal chat message formats to Bedrock's Converse API requirements, including system blocks and tool results.
  • Provided utility functions for converting between serde_json::Value and aws_smithy_types::Document.
  • Included comprehensive unit tests for message conversion, tool configuration, and error mapping.
• src/llm/mod.rs
  • Imported the new bedrock module.
  • Added create_bedrock_provider function to instantiate the BedrockProvider based on configuration.
• src/settings.rs
  • Added bedrock_region and bedrock_cross_region fields to the Settings struct to store AWS Bedrock specific configurations.
• src/setup/README.md
  • Updated the setup README to include AWS Bedrock as a configurable LLM provider, specifying its API key and environment variable.
• src/setup/wizard.rs
  • Updated the setup wizard to include AWS Bedrock as a selectable LLM provider option.
  • Implemented setup_bedrock and setup_bedrock_cross_region functions to guide users through configuring Bedrock region, authentication methods (API key, default credentials, named profiles), and cross-region inference settings.
  • Modified model selection logic to prompt for a Bedrock model ID when Bedrock is the chosen backend.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request introduces support for AWS Bedrock as an LLM provider, including updates to configuration, documentation, and the setup wizard. The core implementation in src/llm/bedrock.rs is well-structured and includes comprehensive tests. I've provided a few suggestions to improve robustness and maintainability, including consolidating duplicated logic as per repository guidelines. Notably, there's a syntax issue in the setup wizard that might cause build failures on stable Rust. Overall, this is a great addition.

[serrrfirat]

Summary

Substantial PR adding AWS Bedrock support via native Converse API. The core implementation — message conversion (with strict user/assistant alternation), tool call handling, Document<->JSON conversion, response extraction, and error mapping — is solid and has good unit test coverage (20+ tests). However, the bearer token (API key) auth path is completely non-functional: AWS_BEARER_TOKEN_BEDROCK is collected and stored but never wired to the AWS SDK client, which does not recognize this env var in the Rust SDK (unlike Python/JS SDKs). This is the auth method advertised as 'simplest' in the wizard. IAM and SSO auth paths work correctly. Additionally, the wizard doesn't persist AWS_PROFILE for named profile auth (inconsistent with other settings), and there's an unsafe u32→i32 cast for max_tokens. Recommend blocking on f-1 (bearer token) and f-2 (profile persistence) before merge.

General findings

[NIT] Settings doc comment doesn't mention bedrock as a valid backend (src/settings.rs:51)

The doc comment on llm_backend field says: 'LLM backend: "nearai", "anthropic", "openai", "ollama", "openai_compatible"' — missing 'bedrock' and 'tinfoil'.

Suggested fix:

Update the comment to: 'LLM backend: "nearai", "anthropic", "openai", "ollama", "openai_compatible", "tinfoil", "bedrock"'.

[zmanian]

Review: AWS Bedrock LLM Provider

The core bedrock.rs implementation is well-structured and handles the Bedrock Converse API correctly. However, there are two blockers preventing merge.

Blocker

1. Bearer token auth is non-functional — The AWS_BEARER_TOKEN_BEDROCK environment variable is read and a Token credential is constructed, but the AWS Rust SDK's Bedrock client doesn't recognize this credential type for the Converse API. Bearer token auth only works with specific AWS services (like CodeCatalyst), not Bedrock. This auth path will silently fail at runtime. Either remove it and document the supported auth methods (env creds, profile, IMDS), or verify it actually works with a test.

2. 36 new crate dependencies with no feature gate — This PR adds 36 new transitive dependencies including aws-lc-sys which requires cmake and a C compiler to build. This significantly increases build times and system requirements for all users, even those not using Bedrock. This must be behind a bedrock feature flag (like the existing libsql feature pattern) so it's opt-in.

High Priority

3. Unsafe u32 as i32 cast — Token usage values are cast with as i32 which silently truncates on overflow. Use i32::try_from(val).unwrap_or(i32::MAX) or similar safe conversion.

4. Fragile error detection — Error handling uses contains("Error:") string matching which is brittle. Match on the actual error variants from the SDK types instead.

Medium Priority

5. Setup wizard integration — The wizard step for Bedrock should only appear when the bedrock feature is enabled, consistent with how other optional backends work.

6. Missing region validation — AWS_REGION defaults to us-east-1 silently. Consider at least logging when the default is used, since Bedrock model availability varies significantly by region.

What's Good

• Clean separation of Bedrock-specific types
• Proper streaming support via Converse Stream API
• Tool use / function calling implementation looks correct
• Good error type mapping

Recommendation

Add a bedrock feature flag to gate the AWS dependencies, remove or fix the bearer token auth path, and this will be close to mergeable. The core implementation quality is solid.

[cgorski]

All PR comments addressed, ready for merge consideration

[ilblackdragon]

Thanks for the work on this, @cgorski! I've picked up your changes and continued them in #713.

The new PR includes your original work plus:

• Merge with latest main (resolved 7 conflicts from the registry-based provider refactor)
• Adapted to new RegistryProviderConfig / string-based backend system
• Added missing struct fields from recent main changes (cache tokens, content_parts)
• Removed non-functional bearer token auth per reviewer feedback from @zmanian and @serrrfirat
• Fixed type mismatches in wizard

You're credited as co-author on all commits. Feel free to review the new PR!

[ilblackdragon]

Thanks for contribution! Taking over in #713 to update to the new format and resolve left over comments.