Installing LLM Gateway
Boomi LLM Gateway is a hosted gateway that lets AI agents, such as Claude Code or Codex, send requests to large language model (LLM) providers like Anthropic and OpenAI without holding a real provider API key.
The agent authenticates to the gateway instead, and the gateway attaches the real provider credential on your behalf.
Request path: Your agent > Boomi AI Gateway > Anthropic / OpenAI / ...
Step 1: Signing in to the admin dashboard
Sign in at https://{YOUR-ORG}.admin.boomi.com (replace with your organization's admin URL) using your Boomi account credentials.
Step 2: Generating a gateway API key
A gateway API key is what your agent uses to authenticate to AI Gateway, instead of a real provider key.
- Open the admin dashboard.
- Go to Security → API Keys.
- Generate a new key. It starts with
brm_llm_. - Copy it, and treat it like a password.
When your agent sends this key, the gateway validates it, removes it, and attaches the real provider credential itself. The provider key never reaches your machine.
Step 3: Choosing an LLM provider
The Organizational Catalog lists every LLM your organization admin has already configured and approved.
- Open the admin dashboard, then go to Organizational Catalog.
- Select the AI tab to view approved providers and the models available under each one.
- Click a provider card for its target host and configuration details.
Step 4: Connecting an agent
Connecting any agent requires the same four values; only where you enter them differs by agent.
| # | Value | What it is | Example |
|---|---|---|---|
| 1 | Gateway address | Where the agent sends requests, in place of the provider's own address | https://{YOUR-GATEWAY-ADDRESS} |
| 2 | Gateway API key | Your brm_llm_ credential from step 2 | brm_llm_abc123... |
| 3 | x-boomi-host header | Which provider the gateway forwards the request to | api.anthropic.com or api.openai.com |
| 4 | x-boomi-agent header (optional) | The label this agent appears under on the usage dashboard | claude-code, codex |
The two headers are optional. Without x-boomi-agent, requests still succeed but appear ungrouped on the dashboard.
Connecting Claude Code
Claude Code reads its connection settings from three environment variables: ANTHROPIC_BASE_URL, ANTHROPIC_API_KEY, and ANTHROPIC_CUSTOM_HEADERS.
The following command configures parameters for a single execution:
env ANTHROPIC_BASE_URL="https://{YOUR-GATEWAY-ADDRESS}" \
ANTHROPIC_API_KEY="brm_llm_YOUR_KEY" \
ANTHROPIC_CUSTOM_HEADERS=$'x-boomi-host: api.anthropic.com\nx-boomi-agent: claude-code' \
claude
To set environment variables for the active shell session, export the variables in the terminal shell. Add export statements to ~/.zshrc for persistent user configurations.
For a permanent, file-based setup, add an env block to ~/.claude/settings.json:
{
"env": {
"ANTHROPIC_BASE_URL": "https://{YOUR-GATEWAY-ADDRESS}",
"ANTHROPIC_API_KEY": "brm_llm_YOUR_KEY",
"ANTHROPIC_CUSTOM_HEADERS": "x-boomi-host: api.anthropic.com\nx-boomi-agent: claude-code"
}
}
Connecting Codex
Codex has no custom-headers variable. Instead, it defines a model provider in its config file and reads the gateway key from a named environment variable.
The base URL must include /v1 (Codex appends /responses itself). Do not store credentials directly within the configuration file.
The following configuration example updates ~/.codex/config.toml:
model_provider = "boomi_gateway"
[model_providers.boomi_gateway]
name = "OpenAI through Boomi AI Gateway"
base_url = "https://{YOUR-GATEWAY-ADDRESS}/v1"
wire_api = "responses"
env_key = "OPENAI_API_KEY"
http_headers = { "x-boomi-host" = "api.openai.com", "x-boomi-agent" = "codex" }
Then export the key:
export OPENAI_API_KEY="brm_llm_YOUR_KEY"
Alternatively, pass configuration parameters via command-line flags using --config.
Connecting a generic agent
Generic AI agents support either the Anthropic dialect or the OpenAI dialect, and expose a base URL setting (sometimes called endpoint or API host) plus an API key setting.
The following matrix specifies base URL format requirements per API format:
| Dialect | Base URL | API key |
|---|---|---|
| Anthropic-style | https://{YOUR-GATEWAY-ADDRESS} | your brm_llm_ key |
| OpenAI-style | https://{YOUR-GATEWAY-ADDRESS}/v1 | your brm_llm_ key |
If the tool supports custom HTTP headers, add x-boomi-host and x-boomi-agent as described in the table in step 4. If it does not, the connection still works; requests are just labeled less precisely on the dashboard.
Step 5: Verifying the connection
Submit a prompt to the agent to verify end-to-end connectivity. To review execution logs and metrics, go to Analytics → Dashboard and verify that requests register under the designated x-boomi-agent label.