Integrations
Works where your agents already run
Anything that speaks the OpenAI or Anthropic API can point at Gatewayz. Each integration below says exactly how far we've tested it.
Coding agents
Claude Code
Tested end to endAnthropic's terminal coding agent, on the native Messages endpoint. Tool use and prompt caching pass through.
export ANTHROPIC_BASE_URL=https://api.gatewayz.ai
export ANTHROPIC_AUTH_TOKEN=YOUR_GATEWAYZ_KEYSetup guide →Cline
Uses the OpenAI-compatible endpointVS Code agent. Choose the OpenAI Compatible provider.
API Provider: OpenAI Compatible
Base URL: https://api.gatewayz.ai/v1
Model ID: anthropic/claude-sonnet-4-6Setup guide →Aider
Uses the OpenAI-compatible endpointTerminal pair programmer. Aider prefixes custom endpoints with openai/.
export OPENAI_API_BASE=https://api.gatewayz.ai/v1
export OPENAI_API_KEY=YOUR_GATEWAYZ_KEY
aider --model openai/anthropic/claude-sonnet-4-6Setup guide →OpenCode
Uses the OpenAI-compatible endpointOpen-source terminal agent, via the @ai-sdk/openai-compatible provider.
"baseURL": "https://api.gatewayz.ai/v1"Setup guide →Continue
Uses the OpenAI-compatible endpointVS Code and JetBrains assistant, with provider set to openai.
"apiBase": "https://api.gatewayz.ai/v1"Codebase indexing needs an embeddings model; namespace it, e.g. openai/text-embedding-3-small.
Setup guide →SDKs
OpenAI SDK
Tested end to endAny language. Change the base URL and use catalog model ids.
base_url="https://api.gatewayz.ai/v1"Setup guide →Anthropic SDK
Native Messages endpointNative Messages, including tool use and prompt caching. The SDK appends /v1/messages itself.
base_url="https://api.gatewayz.ai"Setup guide →Agent platforms
FlashyOS
Default inference providerFlashyOS's published agent CLI ships Gatewayz as its default inference provider: npx @flashyos/agent init --inference gatewayz scaffolds an agent that calls /v1/chat/completions.
What the labels mean
- Tested end to end: we've run it against the production API.
- Uses the OpenAI-compatible endpoint: the tool supports custom OpenAI-compatible providers and the configuration is documented, but we haven't yet run its full workflow (such as an agent editing files through tool calls) end to end.
