Setting Up Model Router
Model Router connects in one of three ways. Most teams start with the hosted proxy — it's the fastest to turn on and requires no provider key management on your side.
Option A: Hosted Proxy (recommended)
The hosted proxy is a gateway Entelligence runs that speaks the same API your Anthropic-compatible tools already expect (/v1/messages). Pointing your agent harness at it is a two-environment-variable change — no SDK upgrade, no code change.
Get your API key
Create one at app.entelligence.ai/settings?tab=api (opens in a new tab) if you don't already have one — the same key used for the CLI and other integrations works here.
Point your tool at the gateway
Set two environment variables in the shell (or CI environment) that launches your coding agent:
export ANTHROPIC_BASE_URL=https://entelligence.ddbrief.com
export ANTHROPIC_AUTH_TOKEN=<your-entelligence-api-key>Claude Code (and any other tool that respects ANTHROPIC_BASE_URL/ANTHROPIC_AUTH_TOKEN) will send every request to the gateway instead of directly to Anthropic. The gateway accepts the standard Anthropic Messages payload, so nothing about how you prompt or use tools changes.
The Model Router dashboard's onboarding screen may show a shortcut command (entelligence router on) that sets these variables for you automatically once it ships to your CLI version. If your installed CLI doesn't recognize it yet, use the two export lines above — they're the underlying mechanism either way.
Confirm it's active
Run a session with your agent, then open the Model Router dashboard (opens in a new tab). It should appear under the Sessions tab within a minute, tagged with the model it was routed to. Every response also carries an x-router-model header if you want to confirm programmatically.
How billing works day-to-day
- Routing draws down a prepaid credit balance. Top up from Model Router → Settings in $25 / $50 / $100 presets, or a custom amount (minimum $5).
- Turn on auto-recharge to top up automatically once your balance drops below a threshold you set — no manual top-ups needed once configured.
- If your balance hits $0, routing stops — the gateway returns an error rather than silently falling back to a direct (unrouted, un-discounted) call to your default provider. You'll see a balance-depleted banner in the dashboard; recharge to resume.
- You're billed 20% of realized savings versus your top-model baseline — if a period doesn't beat that baseline, that period costs $0.
Option B: Bring Your Own Keys (BYOK)
Route across models you already have provider access to, instead of using the hosted gateway.
Open the Model Router onboarding wizard
From Model Router → Get Started in the dashboard.
Connect a provider
Provide an AWS Access Key ID, AWS Secret Access Key, and the AWS Region your Bedrock models are enabled in (e.g. us-west-2).
Credentials are used only to call your provider on Entelligence's behalf for routed requests — traffic is routed across whatever models your account has access to.
Choose a routing mode
Pick Balanced or Eco (see Overview → Routing modes) — you can change this later from the same settings screen you connected the provider in.
BYOK is billed the same way as the hosted proxy — 20% of realized savings versus your top-model baseline, invoiced monthly.
Option C: Bring Your Own Cloud (BYOC)
Run the router entirely inside your own infrastructure. This option is coming soon — contact support@entelligence.ai if this is a requirement for your org.
Switching modes or turning routing off
Both are one click from the Model Router dashboard's Overview tab — no redeploy or config change needed on your side. Turning routing off falls back to calling your configured default model directly (not through the gateway).
Routing decisions today are made globally per org — there's no per-project or per-repo override yet. If you need routing to behave differently for a specific team or workload, talk to us at support@entelligence.ai.
Troubleshooting
| Symptom | Likely cause | Fix |
|---|---|---|
| Requests fail with an auth error | API key invalid, or ANTHROPIC_AUTH_TOKEN not exported in the same shell/process launching your agent | Re-check the key at Settings → API, re-export the variable, and restart your agent session |
| Requests fail with a payment/balance error | Router credits are at $0 | Top up from Model Router → Settings, or enable auto-recharge |
| Nothing shows up in the Sessions tab | Gateway isn't actually receiving traffic | Confirm both env vars are set in the exact process/shell that launches your agent — a new terminal tab won't inherit variables exported elsewhere |
| A session never routes to a cheaper model | Expected in Balanced mode when the task is genuinely complex, or during BYOK if your account lacks access to lower-cost models in the catalog | Check the per-session model breakdown in the dashboard; try Eco mode for lower-stakes work |