Routing

You never choose an upstream provider directly. You use a model name and the gateway routes the request to one of the channels serving that model. What you control, per model, is the routing strategy:

Market mode (default)

The gateway picks the best available channel each request. This is the default and the right choice for almost everyone.

Optionally set price caps: per-channel pricing limits so market mode never routes through a channel above your cap. If every channel serving the model exceeds your caps, the request is rejected - loosen the caps or switch that model back to plain market mode.

Direct mode

Pin a model to specific channels with a channel whitelist:

• Requests for that model go only to whitelisted channels.
• Via MCP set_strategy, setting a new whitelist replaces the previous one; in the console, adding a direct channel adds to the existing whitelist.
• Every whitelisted channel must actually serve the model.

Use direct mode when you need a specific upstream for compliance, latency, or feature reasons. Browse the channel catalog (provider, models served) from the MCP list_channels tool - see Agent Onboarding (MCP).

Where to set strategies

• Console: Routing strategies page.
• MCP: the set_strategy tool (market or direct, optional price_caps, channel_whitelist for direct) and get_strategies to read them back - see the MCP section of https://tokpum.com/llms.txt.

Changes take effect for new requests immediately; there is no restart or save-and-activate step.

What routing does not change

• Billing is per token against your account balance at the model's public prices regardless of channel: https://tokpum.com/api/v1/public/pricing.
• The wire protocol stays OpenAI- or Anthropic-compatible regardless of which channel serves the request.