RouteMyAgentROUTER / AS A SERVICESign in

AFFORDABLE ROUTING. YOUR EXISTING AGENT.

More models.
Same agent.

Keep the coding tools you love.
Bring your NVIDIA key. Rent your own route.

Affordable BYOK routing with idle time paused for Claude Code, Claude Desktop’s Code tab, Codex and OpenCode. Bring your NVIDIA key; provider and client costs stay separate.

Start with 15 minutes free See the route

One trial per account · Your key is encrypted at rest · No automatic renewal

THE ROUTE / ILLUSTRATION

ONE ENDPOINT. YOUR PROVIDER ACCOUNT.

KEEP YOUR AGENT. ADD A SECOND OPINION.

A stronger review.
One MCP connection.

Send a focused debugging question, code review or design decision to your rented model. Keep your own model, tools and files in charge. Prepare your NVIDIA key first, then try 15 minutes of routing free once per account, subject to capacity.

Agent quickstart & NVIDIA key setup ↗

Machine-readable capabilities & live prices · Full agent reference

Stripe prepaid rentals are available. Crypto payments and autonomous purchasing are not live yet. Provider quota and client requirements still apply; new usage time pauses during setup and idle time.

MADE FOR YOUR EXISTING WORKFLOW

Claude Code / Claude Desktop / Codex / OpenCode

Compatible client configuration required. Account requirements still apply.

01 / FROM ZERO TO CONNECTED

Your next session.
Three simple moves.

The server is hosted. The key is yours.
Your agent handles the local setup.

Reserve your route.

Have your NVIDIA account, API key and coding client ready first. Then claim your trial or buy active-use time by the hour. Setup and idle time are paused. Concurrent requests share one usage clock.

TIME FOR YOUR ACTUAL WORK

Bring your NVIDIA key.

Attach your provider key during the active rental. It is encrypted at rest for restart recovery and deleted when your time ends.

ENCRYPTED FOR YOUR RENTAL

Hand your agent the prompt.

Copy your server’s setup prompt. Your agent can configure the correct endpoints and model IDs in the coding client you choose.

YOUR TOOLS, CONNECTED

BUILT AROUND YOUR CONTROL

Your key.
Your control.

We provide the routing infrastructure. You bring your own provider access, choose your model, and decide how long to stay.

The details, without the fine-print hunt ↗

Encrypted recovery

Your NVIDIA key is encrypted in a separate credential store for this rental, excluded from normal ledger backups, and restored after restarts. Disconnecting or expiry deletes it.

No rush while idle

Your purchased time pauses between model requests. Access stops when usage time runs out. A five-minute hold lets you add time for the same model before the slot is released.

A reversible setup

CLI setup uses a separate rental launcher. Claude Desktop uses its Gateway profile with an export and recorded settings for switching back. Both include a connection check.

02 / SIMPLE, TIME-BASED PRICING

Rent the route.
Keep the choice.

GLM 5.3 routing: $3.49 / active-use hour.
Provider and client charges stay separate.

YOUR FIRST CONNECTION

Take it for a spin.

15minutes free

Have your NVIDIA key and a supported coding client ready before starting. Your 15 minutes cover active routing use; setup and idle time are paused.

Choose either model · one trial per Clerk account
Your own reserved slot
The same client setup tools

Start your free trial ↗

WHEN YOU’RE READY TO BUILD

Your routing workspace.

GLM 5.3$3.49 / active-use hour

Kimi K3$4.99 / active-use hour

Prepaid active-use time. Add more when you need it. No automatic renewal.

Choose GLM 5.3 or Kimi K3
Setup and idle time paused
Secure credit checkout with Stripe

Reserve your route ↗Add $5.00 of prepaid credit through Stripe.

Comparing options? See the full cost of AI coding, with hourly examples and when a direct connection may fit better.

New rentals count active routing use, with setup and idle time paused. Concurrent requests count once. Provider usage, quotas, model availability, and applicable terms belong to your NVIDIA account. The trial covers routing only. No refunds for purchased credit or rental time, including unused time, except where required by law. Read the service terms.

03 / GOOD QUESTIONS

Before you
plug in.

What am I paying for?

Prepaid active-use time on the hosted RouteMyAgent router for the model you choose. Model inference comes from your own NVIDIA API account. We do not sell NVIDIA keys, GPU compute or model credits. Your own NVIDIA account supplies inference.

Is this a cheap way to use AI coding models?

It depends on your task and the access you already have. You pay for active-use router hours, plus any NVIDIA or client charges. A direct provider connection may cost less if your client already supports it. Compare the examples and alternatives in our AI coding cost guide before connecting.

How do credit and hourly pricing work?

Add $5.00 in prepaid routing credit, then separately buy a whole-hour block of active routing use: GLM 5.3 $3.49/hour; Kimi K3 $4.99/hour. Remaining credit stays in your balance. Buying credit does not start or renew a rental. New usage time pauses during setup and idle time. Overlapping requests count once. Older wall-clock rentals keep their prior terms. This is router access, not NVIDIA model credits. No automatic renewal; the refund policy below applies.

What should I prepare before the trial?

Create your NVIDIA account and have an authorized API key ready. Have Claude Code CLI, Claude Desktop with Gateway configuration, Codex CLI or OpenCode installed with any account or custom-endpoint access it requires. You can sign in here and read the guide before connecting. The trial provides 15 minutes of active routing use; setup and idle time are paused.

Can I use Claude Desktop and multiple agents?

Yes: choose Claude Desktop’s Code tab and its third-party Gateway setup. Client-managed subagents and parallel local sessions can use the same rented model. They share your NVIDIA quota and usage-time balance; the router does not add a separate worker server or unlock other model tiers. Read the Claude Desktop and Code guide.

Can my own local AI ask a rented model for help?

Yes. Choose “MCP helper (keep my model)” to add RouteMyAgent to an agent that supports Streamable HTTP MCP with a private Bearer header. Your local model stays in charge and can request a review, plan or second opinion. The helper receives only the text your agent sends and returns text; it cannot open your files or run tools. The same rental time and NVIDIA quota apply. A local model API alone or an OAuth-only connector cannot use this connection.

Does this replace a Claude or Codex subscription?

No. The router provides compatible endpoints for supported coding clients. Client account requirements, subscriptions, third-party gateway support, and model capabilities vary. Your setup agent checks the installed client rather than assuming every interface supports custom endpoints.

What happens when the timer reaches zero?

Model access stops when usage time is exhausted. Your slot is held for five minutes so you can add time for the same model. After release, start a new rental and create a new router token. Reattach your NVIDIA key when prompted. No automatic renewal. Older wall-clock rentals keep their original deadline.

Where does my NVIDIA key go?

You attach it through the authenticated portal while your rental is active. It is encrypted at rest for restart recovery, decrypted in memory for upstream requests, and deleted on disconnect or expiry. The separate credential store is excluded from normal ledger backups. Your coding client uses a separate rental token.

Can my agent configure everything for me?

The portal supplies a server-specific setup prompt and configuration references. Give that prompt to your agent, choose the client and model, and enter your router token through a private local prompt. The instructions include preserving your current setup and testing the connection. Never paste keys into an agent conversation.

Can I get a refund?

Purchased credit and rental time are non-refundable, including unused credit and unused time, except where required by law. For payment errors or legally required remedies, contact support@routemyagent.dev. Read the service terms.

What if the provider is rate limited?

Your NVIDIA account’s limits still apply. The router cannot increase its quota. Clients should honor Retry-After and stop resubmitting when usage is exhausted. Failed provider requests do not consume new usage time. Older wall-clock rentals retain their original deadline.

RETURN TO NORMAL / UNDO SETUP

Your usual setup is still yours.

For CLI setup, close the rental child and open your normal launcher. For Claude Desktop, switch to the previous Gateway/connection profile using the saved export and change receipt; closing a window alone does not undo persisted Desktop settings. Use your setup agent’s scoped Undo script for unchanged rental files. Review any edits to shared settings individually, preserving later changes.

Using a browser chat without computer access? It can generate runnable setup and undo files for you to review and run locally. It cannot change your computer or reroute the hosted chat service itself. A local agent with file and shell access can apply the instructions.

Loading server-provided undo instructions…

Preview undo instructions

Undo is available without sign-in and after expiry. It does not refund purchased time or change your credit balance. New usage time pauses whenever no model request is running. Optionally revoke your router token in the signed-in portal; nothing is revoked automatically.

YOUR FIRST CONNECTION

Four steps to your first response.

Choose your rental →

Prepare your key and client; setup does not use new rental time.

01Rent a slotNot started
02Attach NVIDIA keyNeeded
03Create router tokenNeeded
04Set up your clientTest in your client

Before you start: have an authorized NVIDIA account and API key ready, plus Claude Code CLI, Codex CLI or OpenCode installed. Your trial provides 15 minutes of active routing use. Setup and idle time are paused. Signing in does not start model usage.

01 / YOUR RENTALNo active rental

A slot with your name on it.

Usage time remaining

00:00:00

Reserve a slot to open your API endpoint.

Choose your model tier

Each rental permits its selected model only. Renewal keeps the same model and hourly price.

Hours to purchase upfront

New usage time pauses when your agent is idle. Older wall-clock rentals retain their original deadline. No automatic renewal. Requests stop at 00:00. No refunds, including unused time, except where required by law. Service terms.

YOUR ROUTING CREDIT

—

Use your balance to rent or extend.

Have a free-credit code?

Purchased credit and rental time are non-refundable, including unused credit and unused time, except where required by law. Service terms · Privacy

ACTIVE-USE ROUTER HOURS

— / hour

— slots

A slot is reserved for one account. Capacity is checked when you rent.

02 / BRING YOUR OWN KEY

Connect your NVIDIA account.

Not attached

Your key is encrypted at rest for this rental so it survives router restarts. It is decrypted in memory for NVIDIA requests and deleted when you disconnect it or the rental ends.

Get a key from NVIDIA ↗Key setup instructions

Start a rental before attaching a key.

Optional: sends one small request to your rented model using your NVIDIA quota. No automatic retry. This tests the hosted route, not your local coding client.

03 / CREATE YOUR ROUTER TOKEN

Your router tokens.

A router token connects your coding app to your rental. It does not replace your NVIDIA key. Access ends when rental time runs out.

04 / CONNECT YOUR AGENT

Your setup, ready to hand over.

No slot selected

Keep your local AI as the lead and choose MCP helper to add rented-model advice, or choose a client below to use the rental as its main model. Copy or download the matching setup instructions. A local coding agent with file and shell access can create the CLI rental launcher and Undo script, or guide Claude Desktop through a separate saved Gateway configuration with restore instructions. A free browser chat without computer access can generate runnable files for you to review and run locally; it cannot configure your machine or change how its hosted chat is routed. Enter the router token privately during setup for your selected client, never in the prompt.

How do you want to connect?

Choose the same client you use locally.

Agent setup prompt

The prompt contains no secret. Give your local client the router token through AGENTNET_ROUTER_TOKEN. Keep your NVIDIA key out of prompts.

MCP helper endpoint · keep your model

OpenAI-compatible endpoint

Anthropic-compatible endpoint

Models on this server

Rent a slot to see its model catalog.

Client requirements still apply. The setup prompt cannot unlock features restricted by a client account or subscription.

SETUP / CONFIGURATION REFERENCES

Take your configuration with you.

These references come from this server. Follow the selected setup prompt to create a separate rental launcher and a scoped Undo script while preserving your normal configuration. Secrets are never included.

A reference is not a connection test. Follow your setup prompt to preserve existing settings, supply the token privately, and verify the selected client.

RETURN TO NORMAL / UNDO SETUP

Your usual setup is still yours.

For the new launcher setup, close the rental launcher and open your normal coding app; its rental settings apply only to that launched process. To remove its files, use the scoped Undo script your setup agent created. If an older setup or your agent changed saved settings, use the undo prompt and change record to review those edits while preserving later changes.

Loading server-provided undo instructions…

Preview undo instructions

IF SOMETHING STOPS

Get back to your session.

Rental expired / access stopped: Refresh status and check remaining usage time. During the five-minute renewal hold, add time for the same model and reattach your key. After release, start a new rental and create a new router token. Do not retry inference until access is active.
Missing NVIDIA key: Attach it in step 2. Expiry and disconnect clear the attachment; normal server restarts restore it. Your router token is a different credential; it cannot replace the NVIDIA key.
401 / invalid token in your coding client: Use a token created for the current rental, with the exact endpoint in this portal. Set it privately as AGENTNET_ROUTER_TOKEN and apply your selected client’s configuration. Revoke and replace a lost token. A portal sign-in error instead requires signing in again.
429 / rate limit: Honor the response’s Retry-After before trying again. Pause parallel jobs; your NVIDIA account’s quota still applies. Do not repeatedly resubmit or rotate keys to bypass the limit. New usage time is not consumed by a failed provider request.
400 / model or configuration error: Use the rented model and selected client’s prompt/reference below. Read the actual error before changing settings; renting one tier does not enable the other model.
Timeout / 502 / 503 / 504: The request may already have reached the provider. Check your client and refresh portal status before retrying. Reopening a pending credit or rental request keeps its operation ID. There are no automatic paid retries.

Still stuck? Email support@routemyagent.dev with the client, time and error code. Leave out keys, tokens and private prompts.

ACCOUNT / ROUTING CREDIT

Your activity.

Latest 100 entries

Activity	When	Credit change

A Checkout return alone does not credit your account. Confirmed Stripe events update this ledger.

More models.
Same agent.

A stronger review.
One MCP connection.

Your next session.
Three simple moves.

Reserve your route.

Bring your NVIDIA key.

Hand your agent the prompt.

Your key.
Your control.

Encrypted recovery

No rush while idle

A reversible setup

Rent the route.
Keep the choice.

Take it for a spin.

Your routing workspace.

Before you
plug in.

Your usual setup is still yours.

One route. Ready when you are.

Four steps to your first response.

A slot with your name on it.

Connect your NVIDIA account.

Your router tokens.

Your setup, ready to hand over.

Agent setup prompt

Models on this server

Take your configuration with you.

Codex launch arguments

Your usual setup is still yours.

Get back to your session.

Your activity.

Confirm action

More models.Same agent.

A stronger review.One MCP connection.

Your next session.Three simple moves.

Reserve your route.

Bring your NVIDIA key.

Hand your agent the prompt.

Your key.Your control.

Encrypted recovery

No rush while idle

A reversible setup

Rent the route.Keep the choice.

Take it for a spin.

Your routing workspace.

Before youplug in.

Your usual setup is still yours.

Plug intopossibility.

Welcome to RouteMyAgent

Four steps to your first response.

A slot with your name on it.

Connect your NVIDIA account.

Your router tokens.

Your setup, ready to hand over.

Agent setup prompt

Models on this server

Take your configuration with you.

Selected client reference

Codex launch arguments

Your usual setup is still yours.

Get back to your session.

Your activity.

More models.
Same agent.

A stronger review.
One MCP connection.

Your next session.
Three simple moves.

Your key.
Your control.

Rent the route.
Keep the choice.

Before you
plug in.

Plug into
possibility.