Public API address is being connected

Power your AI agents with 280+ modelsup to 99% cheaper than official APIs

Built for Hermes, OpenClaw, Claude Code, and agentic workflows. One key, one unified wallet balance, and full OpenAI/Anthropic SDK compatibility without VPN or multiple subscriptions.

Key issued immediatelyTop up from $5Unused balance refunded
301
Models in catalog
18
Providers
$0.00015 / 1M
Cheapest model
-
Market volume, 24 h
GPT OSS Safeguard 20B$0.00015 / 1M|GLM 4.5 Air$0.0003 / 1M|Qwen 3.8 27B$0.00045 / 1M|MiniMax M2.1$0.000806 / 1M|Gemma 3 12B IT$0.00105 / 1M|Gemma 3 4B IT$0.00105 / 1M|Gemma 4 26B A4B IT$0.001165 / 1M|Qwen3 Coder 30B A3B Instruct$0.0015 / 1M|NVIDIA Nemotron 3 Nano 30B$0.00165 / 1M|NVIDIA Nemotron 3 Super 120B$0.0018 / 1M|Qwen3 32B$0.0018 / 1M|Qwen 3 Next 80b$0.002084 / 1M|Gemma 4 31B$0.0021 / 1M|Ministral 3B$0.00225 / 1M|GPT OSS Safeguard 20B$0.00015 / 1M|GLM 4.5 Air$0.0003 / 1M|Qwen 3.8 27B$0.00045 / 1M|MiniMax M2.1$0.000806 / 1M|Gemma 3 12B IT$0.00105 / 1M|Gemma 3 4B IT$0.00105 / 1M|Gemma 4 26B A4B IT$0.001165 / 1M|Qwen3 Coder 30B A3B Instruct$0.0015 / 1M|NVIDIA Nemotron 3 Nano 30B$0.00165 / 1M|NVIDIA Nemotron 3 Super 120B$0.0018 / 1M|Qwen3 32B$0.0018 / 1M|Qwen 3 Next 80b$0.002084 / 1M|Gemma 4 31B$0.0021 / 1M|Ministral 3B$0.00225 / 1M|
AI Agent InfrastructureSave up to 90%

Built specifically for AI agent builders

Autonomous agents require millions of tokens daily for heartbeats, reflections, and tool runs. llmka.dev provides a unified OpenAI/Anthropic gateway with prices up to 90% cheaper than direct provider APIs and one wallet balance.

Hermes Agent

Self-improving autonomous CLI agent

Autonomous coding & research agent with tool-calling and continuous memory. Fully compatible via OpenAI endpoint.

Full OpenAI compatibility: streaming & tool calls
Prices up to 90% below official vendor tiers
One unified wallet across all model families
~/.hermes/config.yaml
# ~/.hermes/config.yaml
custom_providers:
  - name: llmka
    base_url: https://api.llmka.dev/v1
    key_env: LLMKA_API_KEY
    models:
      claude-sonnet-5: {}
      qwen3.8-flash: {}
      deepseek-v4.1-flash: {}

model:
  default: claude-sonnet-5
  provider: custom:llmka

# ~/.hermes/.env
LLMKA_API_KEY=sk-...
Replace sk-... with your issued key from the dashboardOpenAI / Anthropic Compatible
How it works

Three steps to your first request

No bank cards, no registrations with every provider, no VPN utilities.

01

Sign up and log in

Email or Telegram in one click. An account and balance are created automatically.

02

Top up

USDT (TRC20, BEP20, TON), crypto, or an invoice through a Telegram bot. Minimum payment - $5.

03

Key and requests

Issue a key in the dashboard and send requests to the gateway address handed out with it.

Catalog

Popular models

Prices per million tokens. Full catalog - 280+ models with vendor filters.

129cheaper than official171pricier than official1no reference priceof 301 models in catalog
ModelVendorContextInput / output per 1M24hVs official price

Saving = our price vs the provider’s official price on a basket of 2,000 input + 700 output tokens. Green - we are cheaper, red - pricier. “No reference price” means the model has no official list price to compare with.

Features

What you get

Access infrastructure, not a resale of a single model.

280+ models in one place

GPT, Claude, Gemini, DeepSeek, Qwen, Llama and open weights. Switch models with one line in your request.

One key instead of ten accounts

One key and one balance for every model family instead of registering with each provider.

Crypto payments

USDT (TRC20 / BEP20 / TON), crypto, and invoices through Telegram. No bank cards.

Transparent pricing

Pay strictly for actual tokens at the per-million price. Balance and all wallet write-offs are shown in your dashboard.

Backup routing

Every top-up and write-off goes through the wallet ledger, so unspent funds stay on your balance.

No prompt storage

The gateway does not store request contents: no prompt or response is written to our databases.

Quick start

Connect with one line

Full compatibility with OpenAI and Anthropic SDKs: only base_url changes.

Examples appear once the public API address is connected: the gateway base_url is handed out with the issued key. There is no working endpoint address to print yet.
Economics

Cheaper than managing accounts yourself

Cost comparison for a typical load: 10,000 requests per month, 2,000 input and 700 output tokens per request.

What you doPrice levelComplexities
Directly with each providervendor base rateA bank card for each vendor, separate limits and accounts, manual expense tracking.
Through llmka.devvendor base rate + one balanceOne crypto payment or Telegram invoice, one key, one dashboard with model-level details.

Exact calculation for your load is available in the calculator. The difference from direct payment depends on selected models.

Questions

The essentials, briefly

The gateway follows the OpenAI and Anthropic standards: the base address is handed out together with the issued key, and the rest of your code stays unchanged.
Support

Could not find an answer? Write to us on Telegram - we answer directly.

Telegram supportAll questions
Start in a few minutes

Issue your first key and send a request today

Registration takes less than a minute. Top up from $5. Trial usage is shown line by line in the dashboard.