Kong / Kong/developer.konghq.com
AI Gateway 2.1: Support for routing AWS Bedrock traffic through PrivateLink/VPC endpoints
- Dominant language
- Ruby
- Stars
- 28
- Forks
- 121
- Avg merge
- 1d 4h
- Merged PRs (30d)
- 313
Description
## Overview
AI Gateway 2.1 adds native support for routing AWS Bedrock traffic through PrivateLink/VPC endpoints using a dedicated `vpc_endpoint` configuration.
This replaces the current `upstream_url` workaround, which overrides the entire URL and can interfere with Bedrock path construction and streaming.
## Scope
Document:
* The new Bedrock `vpc_endpoint` configuration.
* When to use it instead of `upstream_url`.
* Support for both **chat/invoke and embeddings**.
* How Kong continues to construct the Bedrock API path while only overriding the connection host.
* SigV4 behavior: signing continues to use the configured AWS region and Bedrock service.
* Behavior/validation when both `vpc_endpoint` and `upstream_url` are configured, based on the final implementation.
## Examples (TBC)
Provide complete **Konnect API** and **kongctl** examples in the Bedrock provider reference page:
```yaml
model:
provider: bedrock
name: anthropic.claude-3-5-sonnet-20241022-v2:0
options:
bedrock:
aws_region: us-east-1
vpc_endpoint: vpce-0123456789abcdef.bedrock-runtime.us-east-1.vpce.amazonaws.com
```
Include examples for both Bedrock inference and embeddings where their configuration differs, confirm the exact schema with engineering
Size: M
Person of contact:
PRD: https://docs.google.com/document/d/11xqivPZX7Kdm9ba13sIPTdHJmwsK1pzksAaNcrBYJM8/edit?tab=t.0#heading=h.mpw3azkpnz6n
Contributor guide
No contributing guide indexed for this repository
Research direction
Start with the Bedrock provider reference page and confirm the final schema and validation behavior with engineering. Document vpc_endpoint versus upstream_url, including SigV4, path construction, chat/invoke, and embeddings. Add complete Konnect API and kongctl examples, and consider the work done when both provider scenarios and the dual-configuration behavior are covered.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- aws
- Domain
- cloud, documentation
- Issue type
- Documentation
- Difficulty
- 3/5
- Estimated time
- 1-2 days
- Activity status
- Active
- Clarity
- Mostly clear
- Newbie friendliness
- 68/100