AWS Bedrock, Tool calling, Helm chart, and Header Rules

We're excited to announce three consecutive Nexus releases that significantly expand the platform's capabilities. These updates introduce AWS Bedrock support for unified foundation model access, tool calling across all major LLM providers, and sophisticated header transformation rules for fine-grained request control.

With Nexus 0.3.2, we've added AWS Bedrock as a fully supported LLM provider, bringing the entire ecosystem of AWS foundation models to your fingertips. This integration provides seamless access to models from industry leaders including AI21, Anthropic, Amazon, Meta, Cohere, DeepSeek, and Mistral—all through a single, unified interface.

Setting up Bedrock is straightforward. Nexus automatically handles AWS authentication through the standard credential chain, supporting everything from IAM roles to local credentials:

[llm.providers.bedrock]
type = "bedrock"
region = "us-east-1"

[llm.providers.bedrock.models."claude-3-5-sonnet"]
model = "anthropic.claude-3-5-sonnet-20241022-v2:0"

[llm.providers.bedrock.models."llama-3-8b"]
model = "meta.llama3-8b-instruct-v1:0"

[llm.providers.bedrock.models."mistral-large"]
model = "mistral.mistral-large-2407-v1:0"

The integration supports full streaming capabilities and maintains compatibility with the OpenAI API format, making it drop-in compatible with existing applications. This means you can leverage AWS's enterprise-grade infrastructure and security while maintaining your current API patterns.

For more details on configuring AWS Bedrock, check out our documentation.

Nexus 0.3.3 introduces tool calling support across all major LLM providers. This powerful feature enables AI models to interact with external systems, APIs, and databases—transforming them from passive responders into active agents that can execute tasks and retrieve real-time information.

Whether you're using Anthropic's Claude, OpenAI's GPT models, Google's Gemini, or AWS Bedrock's foundation models, tool calling works consistently across all providers. It requires no extra configuration or setup.

This unified approach means you can switch between providers without rewriting your tool definitions or changing your application logic. The AI can now fetch data, trigger workflows, and interact with your systems—all while maintaining the security and rate limiting controls that Nexus provides.

Alongside tool calling, we've introduced official Helm chart support, making Nexus deployment in Kubernetes environments as simple as:

helm repo add grafbase https://ghcr.io/grafbase/helm-charts
helm install nexus grafbase/nexus --version 0.1.0

For complete Helm deployment options and configuration, see our Helm chart documentation.

The latest release, Nexus 0.3.4, introduces a header transformation system that gives you fine-grained control over HTTP headers when routing requests to LLM and MCP providers. This feature is essential for enterprise environments where authentication tokens, correlation IDs, and custom metadata need careful management.

The new header rules support four core operations that can be combined to handle any routing scenario:

# Forward specific headers from the client
[[llm.providers.openai.headers]]
rule = "forward"
name = "X-Request-Id"

[[llm.providers.openai.headers]]
rule = "forward"
name = "X-Correlation-Id"

# Insert new headers with static values
[[llm.providers.openai.headers]]
rule = "insert"
name = "X-Source"
value = "nexus-gateway"

[[llm.providers.openai.headers]]
rule = "insert"
name = "Authorization"
value = "Bearer ${API_TOKEN}"

# Remove sensitive headers before forwarding
[[llm.providers.openai.headers]]
rule = "remove"
name = "X-Internal-Token"

# Rename headers for compatibility
[[llm.providers.openai.headers]]
rule = "rename_duplicate"
name = "X-Client-Token"
rename = "Authorization"

Beyond individual header management, Nexus supports powerful regex patterns to match multiple headers at once:

# Forward all tracing headers with a single rule
[[llm.providers.anthropic.headers]]
rule = "forward"
pattern = "^X-Trace-"

# Remove all debug headers
[[llm.providers.anthropic.headers]]
rule = "remove"
pattern = "^X-Debug-"

# Forward correlation headers
[[llm.providers.anthropic.headers]]
rule = "forward"
pattern = "^X-Correlation-"

This level of control is particularly valuable in multi-tenant environments where different clients may require different authentication schemes, or when integrating with legacy systems that expect specific header formats.

Learn more about header transformation rules in our documentation.

These releases represent a significant step forward in Nexus's evolution as a AI gateway. With AWS Bedrock support, you get access to a vast ecosystem of foundation models. Tool calling transforms your AI from a conversational partner into an active assistant. And header transformation rules provide the enterprise-grade control needed for production deployments.

We're committed to making Nexus the most flexible and powerful AI router available. Stay tuned for more updates as we continue to expand provider support, enhance security features, and improve the developer experience.

To get started with these new features, update to the latest Nexus:

docker pull ghcr.io/grafbase/nexus:stable

For configuration examples and migration guides, check out our documentation and join the discussion on our Discord community.