Gateway Routing

The Prism plugin supports two operating modes that control how your Claude API requests are handled.

Telemetry-only mode

Gateway routing: disabled

Your Claude API requests go directly to Anthropic. The plugin only captures telemetry (OTEL logs, metrics, traces) and prompt text — sent to the Prism ingest service in the background.

What you get:

• PRISM scoring and coaching
• Token and cost tracking
• Waste and throttle detection
• All plugin commands work normally

What you don’t get:

• Budget enforcement (soft/hard caps)
• Guardrails (DLP, PII detection, prompt injection guard)
• Request-level access control

Full governance mode

Gateway routing: enabled (default)

Your Claude API requests route through the Optra gateway before reaching Anthropic. The gateway applies policies, then forwards the request.

What you get (in addition to telemetry):

• Budget enforcement — soft warnings and hard blocks at spend limits
• Guardrails — DLP, PII detection, content filtering, prompt injection guard
• Model access control — restrict which models developers can use
• Request logging — full request/response capture at the gateway level

How it works

When gateway routing is enabled, the plugin sets two environment variables on session start via CLAUDE_ENV_FILE:

ANTHROPIC_BASE_URL={gateway_url}
ANTHROPIC_CUSTOM_HEADERS=X-Gateway-Api-Key: {api_key}
x-prism-source: claude-code

The gateway URL is resolved automatically from your API key via the config endpoint — you don’t need to configure it manually.

Claude Code picks these up and routes all API calls through the gateway instead of directly to api.anthropic.com.

Toggling

Switch between modes:

/prism:status toggle

This flips enableGateway in your config and prompts you to restart Claude Code for the change to take effect.

You can also edit the legacy config directly:

{
  "enableGateway": false
}

When to use each mode

Scenario	Recommended mode
Personal use, just want insights	Telemetry-only
Team with budget limits	Full governance
Enterprise with compliance requirements	Full governance
Debugging latency issues	Telemetry-only (removes gateway hop)
Air-gapped or restricted network	Telemetry-only