Skip to content
Bankr Support home
InboxAsk a human

LLM Gateway & Credits

The LLM Gateway is an OpenAI-compatible proxy that routes to multiple AI providers. It's how Bankr pays for compute — and how your agent can fund itself: fees earned from token trading accumulate in your agent's wallet and can pay for LLM inference through the gateway automatically. Your agent earns fees, fees pay for compute, the agent keeps running.

Available Models

Max Mode gives you access to premium models from Anthropic (Claude), OpenAI (GPT), Google (Gemini), xAI (Grok), Moonshot, Alibaba, and more. When you aren't in Max Mode, the agent runs on a fast base model. The current model lineup and pricing are always live at bankr.bot/llm — that page is the source of truth, since models are added, discounted, and retired regularly.

The web terminal supports all gateway models. The CLI's --model flag accepts a curated subset that expands with each release — run bankr llm models for the current list, and update the CLI (bun install -g @bankr/cli) if a newly launched model isn't accepted yet.

The model list is read live from the gateway, so a newly added model, a pricing change, or a model being taken offline reaches Max Mode within about a minute — no release required. If a model is withdrawn while you're mid-conversation, the run is still billed at that model's last published rate. Requests to retired model names are rejected before routing — if that happens, pick a current model from bankr llm models.

Managing Credits

"add $25 in LLM credits"
"what's my credit balance?"

CLI:

bankr llm credits           # check balance
bankr llm credits add 25    # add $25
bankr llm credits auto --enable   # auto top-up

Model Discounts

The gateway supports time-bounded per-model discounts — new models often launch with intro pricing, and existing models get periodic promo rates. Each discount has its own validity window, and your effective pricing (including any discounts you qualify for) is always shown live at bankr.bot/llm. Check there for what's active right now.

What Happens When Credits Run Out?

Every credit check runs against your effective balance: spendable credit minus usage that's been metered but not yet settled. A run that just finished can still be pending against it, so your effective balance can be lower than the raw number you see.

  • Gateway API requests — rejected once your effective balance reaches zero.

  • Max Mode in chat — a new message is turned away with a top-up prompt (on X it falls back to the base model). A message already in flight stops as soon as its spend reaches your balance, telling you what it used and which transactions it had already submitted. Progress stays in the thread — top up and ask Bankr to continue.

  • Automations — skipped runs are recorded in the automation's history, and 3 consecutive credit failures pause the automation.

Usage your balance can't cover isn't written off — it stays owed and settles out of your next top-up. Enable auto top-up (bankr llm credits auto --enable) to avoid running dry mid-task.

Building Against the Gateway API?

The gateway passes Claude requests through natively (prompt caching preserved end to end — cache reads and writes both appear on your usage, with writes billed at the cheaper cache-write rate). API usage, endpoints, and integration details live in the LLM Gateway docs.

For full details, see LLM Gateway, Max Mode, and Credits on docs.bankr.bot.