GitHub Copilot Chat Model Provider for Amazon Bedrock

A VS Code extension that registers Amazon Bedrock as a model provider for GitHub Copilot Chat (the built-in VS Code chat), so you can drive Copilot Chat with Amazon Bedrock–hosted models — Claude, Llama, Mistral, Qwen, and more.

It contributes two separate Copilot Chat model providers, because the native and Mantle access paths are genuinely different Amazon Bedrock endpoints — different regions, different auth details, and different model coverage:

• Amazon Bedrock (native Converse API) — the full Amazon Bedrock foundation model catalog, including all Anthropic Claude models, across 18 AWS regions.
• Amazon Bedrock Mantle (OpenAI-compatible + Anthropic Messages API) — open-weight models (DeepSeek, Mistral, Qwen, GLM, Nemotron, MiniMax, Kimi, Gemma, gpt-oss) plus the subset of current Claude models Mantle supports, across 13 AWS regions.

Both show up together in the Copilot Chat model picker; enable either or both independently.

• Keep code and prompts in your AWS account for stronger governance
• Choose your AWS region to align with residency and compliance requirements
• Streaming + tool calling for responsive coding workflows
• Multi-region support across 18 AWS regions (native) / 13 AWS regions (Mantle)

Why This Extension

• Compliance-first architecture: prompts, code context, and responses stay within your AWS account boundary.
• Data residency control: select the AWS region your team is allowed to use and keep traffic there.
• Enterprise-ready access model: works with existing AWS credentials, profiles, and IAM controls.
• No model lock-in: use multiple Amazon Bedrock model families from one Copilot Chat workflow.
• Built for developer UX: streaming responses, tool calling, and model switching in the standard chat UI.

Supported Model Families

OpenAI

• gpt-oss-20b, gpt-oss-120b
• Safeguard variants: gpt-oss-safeguard-20b/120b

Google

• Gemma 3: 4b, 12b, 27b variants

Mistral

• magistral-small-2509
• mistral-large-3-675b-instruct
• Ministral: 3b, 8b, 14b variants
• Voxtral: mini-3b, small-24b variants

Qwen

• General: qwen3-32b, qwen3-235b, qwen3-next-80b
• Vision: qwen3-vl-235b (multimodal)
• Coding: qwen3-coder-30b/480b

DeepSeek

• v3.1

Nvidia

• nemotron-nano-9b-v2, nemotron-nano-12b-v2

Others

• MoonshotAI: kimi-k2-thinking
• Minimax: minimax-m2
• ZAI: glm-4.6

Prerequisites

Authentication options:

1. API key mode (optional):
   • Use an Amazon Bedrock API key from the Amazon Bedrock console
2. AWS credentials mode (optional):
   • Use AWS credentials/profile available to VS Code (env vars, ~/.aws/credentials, SSO, etc)
   • You can also set aws-bedrock.awsProfile
3. VS Code: Version 1.104.0 or later

The extension includes a built-in profile picker, status headers, and connection test to help you debug authentication setup.

Installation

From VS Code Marketplace

1. Open VS Code
2. Go to Extensions (Cmd+Shift+X)
3. Search for "GitHub Copilot Chat Model Provider for Amazon Bedrock"
4. Click Install

From Source

1. Clone this repository:

   git clone https://github.com/easytocloud/bedrock-vscode-chat.git
   cd bedrock-vscode-chat
   

2. Install dependencies:

   npm install
   

3. Compile the extension:

   npm run compile
   

4. Press F5 to open a new VS Code window with the extension loaded

Setup & Quick Start

Native Amazon Bedrock and Mantle each have their own settings, since they're independent providers with independent AWS endpoints, regions, and authentication details — but both are configured from one place.

Quick Start: The Amazon Bedrock (Copilot Chat) Sidebar

Click the Amazon Bedrock (Copilot Chat) icon in the Activity Bar to open the configuration panel — an accordion with a section per provider (and one for chat behavior / model metadata), showing live status, "Test Connection" actions, and real-time AWS profile discovery from ~/.aws/config, all in one drawer instead of a chain of Command Palette menus.

The same panel opens via the gear icon next to the model picker in the Copilot Chat panel, or via the Show Configuration command from the Command Palette (Cmd+Shift+P / Ctrl+Shift+P).

1. Native Amazon Bedrock (Converse API)

In the Native (Converse API) section:

1. AWS Profile — optional named profile (leave blank for the default AWS credential chain). The dropdown auto-discovers profiles from ~/.aws/config and ~/.aws/credentials.
2. Region — choose from native Amazon Bedrock's 18 supported regions (default us-east-1)
3. Test Connection — run a real credential check to verify your setup works

Native Amazon Bedrock always authenticates with AWS credentials (env vars, ~/.aws/credentials, SSO, etc.) — there's no API key option here.

2. Mantle (OpenAI-compatible + Anthropic Messages)

In the Mantle (OpenAI-compatible) section:

1. Authentication Method — choose API Key (simpler) or AWS Credentials
   • API Key: paste your key from the Amazon Bedrock console directly into the panel. You'll also be prompted automatically on first use. Keys are stored in VS Code's SecretStorage.
   • AWS Credentials: uses AWS Signature V4 with your existing credentials; optionally set a specific profile (auto-discovered from ~/.aws/config)
2. Region — choose from Mantle's 13 supported regions (a strict subset of native Amazon Bedrock's — default us-east-1)
3. Test Connection — run a real credential check to verify your setup works

Configure via Settings

Both providers' settings can also be set directly:

{
  "aws-bedrock.region": "us-west-2",               // native Amazon Bedrock region
  "aws-bedrock.awsProfile": "my-profile",          // native Amazon Bedrock AWS profile

  "aws-bedrock.mantleRegion": "us-east-1",         // Mantle region (smaller region list than native)
  "aws-bedrock.mantleAuthMethod": "awsCredentials", // or "apiKey"
  "aws-bedrock.mantleAwsProfile": "my-profile"      // optional
}

Configure Model Visibility (Optional)

Show/hide specialized models (like safeguard variants), and optionally hide open-weight models from the native picker when they're also available via Mantle:

{
  "aws-bedrock.showAllModels": true,               // default: true
  "aws-bedrock.hideMantleModelsFromNative": false  // default: false
}

Usage

Using in Chat

1. Open GitHub Copilot Chat (Cmd+Shift+I / Ctrl+Shift+I)
2. Click the model picker (top of chat panel)
3. Select an Amazon Bedrock model (e.g., "OpenAI GPT OSS 120B")
4. Start chatting!

Using with Copilot Chat

1. In any editor, use @workspace or other chat participants
2. The model picker will include Amazon Bedrock models
3. Select an Amazon Bedrock model for your conversation

Example Chat

You: What are the key features of Rust's ownership system?

Assistant (via Amazon Bedrock): [Streams response in real-time...]

Configuration

Settings

Settings are grouped into sections in VS Code's Settings UI (search aws-bedrock): Amazon Bedrock (Copilot Chat), › Native (Converse API), › Mantle (OpenAI-compatible), › Chat Behavior, and › Model Metadata.

Setting	Type	Default	Description
aws-bedrock.showAllModels	boolean	true	Show all models including specialized variants (applies to both providers)
aws-bedrock.logLevel	string	info	Output channel verbosity: verbose, info, warning, error, or none (applies to both providers)
aws-bedrock.requestTimeout	number	120000	Max time (ms) to wait for a response before aborting; 0 disables. Streaming requests only time out waiting for the first response (applies to both providers)
aws-bedrock.enableNative	boolean	true	Register the native Amazon Bedrock (Converse API) provider
aws-bedrock.awsProfile	string	empty	Optional AWS profile for native Amazon Bedrock
aws-bedrock.region	string	us-east-1	AWS region for native Amazon Bedrock (18 supported regions)
aws-bedrock.hideMantleModelsFromNative	boolean	false	Hide open-weight models from the native list when also available via Mantle
aws-bedrock.enableMantle	boolean	true	Register the Mantle provider
aws-bedrock.mantleAuthMethod	string	apiKey	Mantle auth mode: apiKey or awsCredentials
aws-bedrock.mantleAwsProfile	string	empty	Optional AWS profile for Mantle when using AWS credentials
aws-bedrock.mantleRegion	string	us-east-1	AWS region for Mantle (13 supported regions — a subset of native's)
aws-bedrock.sendTools	boolean	true	Send tool definitions to the model
aws-bedrock.emitPlaceholders	boolean	true	Emit placeholder text while waiting
aws-bedrock.enablePromptCaching	boolean	true	Insert Amazon Bedrock cachePoint checkpoints for native Converse requests to Claude/Nova models
aws-bedrock.assumeLongContextClaudeModels	boolean	true	Report a 1M-token context window for Claude models known to support it (Sonnet 4, Sonnet 4.6, Sonnet 5, Opus 4.6+)
aws-bedrock.modelMetadataSource	string	none	Metadata source for token/capability info: none (built-in heuristics only, no network calls) or litellm
aws-bedrock.modelMetadataUrl	string	default URL	External metadata registry URL (used when source is litellm)
aws-bedrock.modelMetadataCacheHours	number	24	Cache duration for external metadata (used when source is litellm)

Supported Regions

Native Amazon Bedrock (aws-bedrock.region) — 18 regions:

us-east-1 (default), us-east-2, us-west-1, us-west-2, ca-central-1, eu-west-1, eu-west-2, eu-west-3, eu-central-1, eu-north-1, eu-south-1, ap-south-1, ap-northeast-1, ap-northeast-2, ap-southeast-1, ap-southeast-2, ap-southeast-3, sa-east-1

Mantle (aws-bedrock.mantleRegion) — 13 regions, a strict subset of native's (missing us-west-1, ca-central-1, eu-west-3, ap-northeast-2, ap-southeast-1):

us-east-1 (default), us-east-2, us-west-2, eu-west-1, eu-west-2, eu-central-1, eu-north-1, eu-south-1, ap-south-1, ap-northeast-1, ap-southeast-2, ap-southeast-3, sa-east-1

Commands

All commands appear in the Command Palette under the Amazon Bedrock (Copilot Chat) category.

Command	Description
Show Configuration	Open the Amazon Bedrock (Copilot Chat) configuration drawer (same panel as the Activity Bar icon)
Manage Native Provider (Converse API)	Opens the same configuration drawer for the native provider
Manage Mantle Provider (OpenAI-compatible)	Opens the same configuration drawer for the Mantle provider
Clear Mantle API Key	Remove the stored Amazon Bedrock API key used by Mantle
Show Logs	Open the extension output channel

Architecture

This extension registers two VS Code LanguageModelChatProvider implementations, because the native and Mantle access paths are genuinely different Amazon Bedrock endpoints with different region footprints, auth details, and model coverage.

Key Components

• NativeBedrockProvider (src/provider.ts): Native Bedrock (Converse API) provider, using @aws-sdk/client-bedrock-runtime
• MantleProvider (src/mantleProvider.ts): Mantle provider — dispatches to Chat Completions (OpenAI-compatible) for most models, and to the Anthropic Messages API for the subset of Claude models Mantle supports
• Mantle Messages API client (src/mantleMessages.ts): Request/response conversion and SSE streaming for Mantle's /anthropic/v1/messages
• Dynamic Model Discovery: Fetches available model catalogs from Amazon Bedrock APIs for both providers
• Streaming Support: Processes SSE (Server-Sent Events) for real-time responses on all three API paths
• Tool Calling: Buffers and parses streaming tool calls for function calling support

API Endpoint Formats

https://bedrock-mantle.<region>.api.aws/v1                    # Mantle: models list, Chat Completions
https://bedrock-mantle.<region>.api.aws/anthropic/v1/messages # Mantle: Anthropic Messages API (Claude)

Native Bedrock uses the AWS SDK (@aws-sdk/client-bedrock / @aws-sdk/client-bedrock-runtime) rather than a raw HTTP endpoint.

Model Capabilities

Tool Calling Support

Models with function calling capabilities:

• gpt-oss-120b
• mistral-large-3-675b-instruct
• magistral-small-2509
• deepseek.v3.1
• qwen3-235b and larger models
• qwen3-vl-235b (vision + tools)
• All Anthropic Claude models (native Amazon Bedrock)
• Amazon Nova, Cohere Command R/R+, AI21 Jamba (native Amazon Bedrock)

VS Code's Agent mode only shows models that report tool-calling support in the model picker — a model without it is hidden there entirely (though still usable in plain Ask-mode chat). If a model you expect to see is missing from Agent mode, it's most likely a tool-calling capability gap; please open an issue.

Vision Support

Models with multimodal (image) input:

• Models from API-key mode: based on model naming and API behavior
• Models from Converse API mode: based on Amazon Bedrock's reported input modalities

Notes on Capability Metadata

• Token limits + initial capabilities: By default the extension uses built-in heuristics only, with no external network calls. It can optionally use an external model metadata registry (litellm's public JSON) for more accurate limits on non-Claude Mantle models — set aws-bedrock.modelMetadataSource to litellm to enable it (also configurable via aws-bedrock.modelMetadataUrl and aws-bedrock.modelMetadataCacheHours). For Claude specifically, aws-bedrock.assumeLongContextClaudeModels (on by default) reports the correct 1M-token window for the specific models that support it, without needing the external registry.
• Native Amazon Bedrock models: vision is derived from ListFoundationModels input modalities (reliable). Tool support is verified on-demand by attempting a tool-enabled request and caching whether the model accepts tool config (this overrides external metadata if runtime behavior differs).
• Mantle models: /v1/models does not include full tool/vision/token metadata, so the extension uses external metadata when enabled, plus runtime probing (tools) as a safety net. Claude models on Mantle are routed to the Anthropic Messages API automatically; models with zero Mantle support (neither Chat Completions nor Messages) are excluded from the Mantle picker entirely.
• Duplicate names: if two models in the same provider's list would otherwise show an identical label, the extension appends the raw technical model ID to keep every entry uniquely selectable.

Code Specialization

Models optimized for coding:

• qwen3-coder-30b-a3b-instruct
• qwen3-coder-480b-a35b-instruct

Reasoning/Thinking

Models with enhanced reasoning:

• kimi-k2-thinking

Troubleshooting

API Key Issues

Problem: "Invalid API key" error

Solution:

1. Verify your API key in the Amazon Bedrock console
2. Open the Amazon Bedrock (Copilot Chat) configuration drawer → Mantle section → "Clear API Key"
3. Re-enter your API key

Model Not Available

Problem: "Model not available in region" error

Solution:

• Not all models are available in all regions
• Try changing to us-east-1 (widest availability)
• Check Amazon Bedrock Model Availability

Rate Limiting

Problem: "Rate limit exceeded" error

Solution:

• Wait a few moments and try again
• Consider using smaller models for testing
• Check your Amazon Bedrock quotas in the AWS console

Connection Issues

Problem: Network or timeout errors

Solution:

• Check your internet connection
• Verify firewall/proxy settings allow access to *.api.aws
• Ensure the selected region is accessible from your location

Development

Building from Source

# Install dependencies
npm install

# Compile TypeScript
npm run compile

# Watch mode for development
npm run watch

# Run linting
npm run lint

Or use the Makefile shortcuts:

make install
make compile
make watch
make lint

Debugging

1. Open the project in VS Code
2. Press F5 to launch Extension Development Host
3. Set breakpoints in source files
4. Test the extension in the new window

Project Structure

bedrock-vscode-chat/
├── src/
│   ├── extension.ts           # Extension entry point; registers both providers
│   ├── provider.ts             # NativeBedrockProvider (Converse API)
│   ├── bedrockNative.ts        # Native Amazon Bedrock Converse API client
│   ├── mantleProvider.ts        # MantleProvider (Chat Completions + Messages dispatch)
│   ├── mantleMessages.ts        # Mantle Anthropic Messages API client
│   ├── externalModelMetadata.ts # External model metadata loader
│   ├── regions.ts               # AWS region lists (native + Mantle, single source of truth)
│   ├── types.ts                # TypeScript type definitions
│   └── utils.ts                # Utility functions
├── package.json                # Extension manifest
├── tsconfig.json               # TypeScript configuration
├── assets/
│   ├── icon.svg                # Source icon (editable)
│   └── icon.png                # Extension icon (128x128)
├── docs/
│   ├── CONTRIBUTING.md         # Development guide
│   ├── QUICKSTART.md           # Quick start for developers
│   ├── TESTING.md              # Testing guide
│   └── dev/
│       └── MODEL_STRATEGIES.md # Model capability detection strategy
├── config/
│   ├── Makefile                # Build automation
│   ├── esbuild.js              # Bundler configuration
│   ├── eslint.config.js        # Linter configuration
│   └── tsconfig.json           # TypeScript configuration
├── README.md                   # This file

Contributing

Contributions are welcome! See CONTRIBUTING.md for detailed development guidelines.

Quick start for contributors:

1. Fork and clone the repository
2. Install dependencies: npm install
3. Compile: npm run compile
4. Press F5 to launch Extension Development Host
5. See docs/CONTRIBUTING.md for testing, logging, and publishing guidelines

Key development notes:

• Publisher name: easytocloud (lowercase)
• Use Output Channel for logging, not console.log
• The extension is bundled with esbuild (out/extension.js) — node_modules is excluded from the VSIX
• Test in both F5 mode and installed VSIX
• Use rsvg-convert for icon generation

Resources

• Amazon Bedrock Documentation
• VS Code Language Model API
• OpenAI API Reference
• Contributing Guide - Detailed development documentation

License

MIT License - See LICENSE file for details

Credits

• Project Lead: easytocloud
• Development Assistant: GitHub Copilot

Acknowledgments

Inspired by the HuggingFace extension for GitHub Copilot Chat.

Support

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Amazon Bedrock: AWS Support

---

Amazon Bedrock, AWS, and Amazon Web Services are trademarks of Amazon.com, Inc. or its affiliates. This extension is an independent, community-built GitHub Copilot Chat model provider and is not affiliated with, endorsed by, or sponsored by Amazon.com, Inc. or its affiliates.

Version: 0.6.0
Status: Production Last Updated: July 11, 2026