KodaCode (Koda): bring your own Claude, GPT and DeepSeek with a RuAPI key — no VPN, no foreign card
Koda is a Russian AI coding assistant: the Koda Desktop app, a plugin for VS Code and JetBrains, and the Koda CLI terminal tool. Out of the box it runs on its own koda-base and koda-pro models, and external models spend plan credits. On top of that, Koda lets you connect any OpenAI-compatible model with your own key — Koda does not bill those requests. Point it at https://www.ruapi.ai/v1 with a RuAPI key, and Claude, GPT, Gemini and DeepSeek join Koda’s model list, billed per token and paid in USDT.
Updated:
from openai import OpenAI
client = OpenAI(
api_key="sk-...",
base_url="https://www.ruapi.ai/v1",
)
resp = client.chat.completions.create(
model="deepseek-flash",
messages=[{"role": "user", "content": "Hello"}],
)
print(resp.choices[0].message.content)Before setting up Koda, test the key in a terminal (cURL tab): the example calls deepseek-flash, which works right after sign-up. Koda then needs the same two values, Base URL https://www.ruapi.ai/v1 and the API key, plus a model ID copied from the pricing page.
When you need your own key in Koda
Credits for external models run out fast.
Claude, GPT and other external models in Koda deduct RC at per-model multipliers. On the Free plan their allowance drops to 0 RC from the second month, and the paid Pro plan has none from the start — only add-on packs (kodacode.ru/tariffs, 3 Nov 2026). A model connected with your own key spends no Koda credits at all: you pay only the provider that issued the key (docs.kodacode.ru, 3 Nov 2026).
You need a specific model, not whatever is on the list.
The Koda team decides which external models it offers. Through RuAPI you pick a model by ID: claude-sonnet-5-5 or claude-opus-5-5 for hard refactoring, gpt-5.4 for review, deepseek-flash for cheap bulk edits — and keep several such connections in Koda on one key.
You want to see what each request costs.
An RC is a notional unit, roughly one average request, and one agent task usually takes 10–15 requests (docs.kodacode.ru, 3 Nov 2026). In Usage Logs in the RuAPI Console every request shows the model, the tokens and the amount in dollars. The balance is topped up in USDT, with no subscription.
How to connect RuAPI to Koda: 3 steps
Checked against Koda CLI and the VS Code plugin 1.3.0 (30 Oct 2026). Regular Koda accounts can add their own models; business and SaaS accounts are limited to Koda’s managed models (docs.kodacode.ru, 3 Nov 2026). RuAPI is an independent service, not affiliated with Koda LLC: you connect it through Koda’s standard bring-your-own-model feature. Koda’s documentation is in Russian, so Russian field labels are given in brackets.
Sign up for RuAPI and top up
Sign-up at www.ruapi.ai works with email, Google or Yandex. Balances are funded in USDT, 5 USDT or more, over one of four networks (TRC20, BEP20, TON, Polygon); USDC is also fine. SBP lets you buy USDT for rubles, and docs.ruapi.ai/en/buy-usdt breaks it down step by step. A foreign bank card is another option.
To test the Koda connection, the starting $0.05 that new accounts usually get is enough. On that trial balance claude-haiku-5-5, deepseek-flash and gpt-6-luna respond, and the remaining models open with the first top-up.
Create a key for Koda
Console → API Keys → "Create API key". Koda’s agent mode fires off one request after another by itself, so put a spending cap on the key: switch "Unlimited" off under Credit limit and set a sum. A separate key also shows in Usage Logs how much Koda alone is spending.
Add the model in Koda
- Koda Desktop: Settings → Models tab («Настройки» → «Модели»; visible once you sign in to a Koda account). Model name («Имя модели») — anything, e.g. RuAPI Sonnet 5.5. Base URL — replace the prefilled https://api.openai.com/v1 with https://www.ruapi.ai/v1. API key («API-ключ») — your sk-… key. Model ID («Идентификатор модели») — the ID from the pricing page, e.g. claude-sonnet-5-5; once Base URL and key are filled in, Koda suggests a list of models. Click Add model («Добавить модель»).
- Koda CLI: type /model and pick "[+] Add custom model (OpenAI-compatible)" at the bottom of the list. Fill in Name, Model ID, Base URL (https://www.ruapi.ai/v1) and API key. The CLI checks the ID against /models and saves the connection to ~/.config/koda/credentials.json; in /model it shows up marked (custom). According to the docs, Koda Desktop also keeps its models through the CLI.
- VS Code and JetBrains plugin: your own models go into ~/.koda/config.yaml (%USERPROFILE%/.koda/config.yaml on Windows). Add an entry to the models list with name: RuAPI DeepSeek Flash, provider: openai, model: deepseek-flash, apiBase: https://www.ruapi.ai/v1, apiKey: sk-…, roles: [chat, edit, agent, plan] and capabilities: [tool_use]. The roles are listed explicitly so the model handles chat, edits, agent mode and planning but not autocomplete; capabilities: [tool_use] lets it call tools in agent mode. Keep the Koda Models entry — those are the built-in models.
- The “Context window, tokens” field in Koda Desktop and the CLI is optional. RuAPI does not report window sizes in its model list, so with the field empty Koda assumes a 200,000-token window. deepseek-flash and claude-sonnet-5-5 have a 1,000,000-token window: enter a larger number if the agent needs a long history, or leave the field empty — the lower the limit, the sooner Koda compresses history and the cheaper each request.
RuAPI address and keys: docs.ruapi.ai/en/quickstart · tool calling: docs.ruapi.ai/en/function-calling · error codes: docs.ruapi.ai/en/errors
What to enter in Koda
Koda Desktop → Settings → Models
Base URL: https://www.ruapi.ai/v1
API key: sk-…
Model ID: claude-sonnet-5-5
Koda CLI → /model → "[+] Add custom model"
Base URL: https://www.ruapi.ai/v1
API key: sk-…
Model ID: deepseek-flash
Plugin → ~/.koda/config.yaml
provider: openai
apiBase: https://www.ruapi.ai/v1
apiKey: sk-…
model: gpt-5.4
roles: [chat, edit, agent, plan]
capabilities: [tool_use]
Payment
USDT / USDC from 5 USDT
foreign-issued cards
no subscription
Model prices for Koda
The models developers reach for most often; prices are per 1M tokens in USD, straight from the pricing page. Your RuAPI balance is charged only for the tokens Koda actually sent and received; no Koda credits are used.
| model | input $/M | output $/M |
|---|---|---|
| deepseek-flash | $0.17 cache read $0.0120 | $1.44 |
| gpt-5.4-mini | $0.90 cache read $0.0900 | $5.40 |
| claude-haiku-5-5 | $0.12 cache read $0.0120 >100K: $0.60 | $0.60 >100K: $3.00 |
| deepseek-v4-pro | $1.58 cache read $0.0600 | $4.75 |
| gpt-5.4 | $3.00 cache read $0.30 >272K: $6.00 | $18.00 >272K: $27.00 |
| claude-sonnet-5-5 | $2.40 cache read $0.12 | $12.00 |
| claude-opus-5-5 | $4.80 cache read $0.24 | $24.00 |
Input means fresh prompt tokens, “cache read” means prompt tokens the model got from cache, output means reply tokens. The Koda agent resends the whole accumulated history every time, so caching makes a real difference to the bill. GPT and DeepSeek put an unchanged request prefix into cache without any setup. We found no cache_control marks in the Koda CLI 1.3.0 code (checked 3 Nov 2026), so for Claude in Koda Desktop and Koda CLI RuAPI adds them itself when the next agent step comes within 5 minutes (docs.ruapi.ai/en/prompt-caching).
The “>100K” line in grey is the vendor’s price above the prompt-length threshold; a request that crosses it is billed at that price in full.
Koda’s models or your own RuAPI key
| Your model via RuAPI | Koda models on a plan | |
|---|---|---|
| How you pay | USDT balance, billed per token, no subscription | Free, Pro, Pro+ and Ultra plans in rubles, plus RC packs |
| Which models | Any model in the RuAPI catalog by ID: Claude, GPT, Gemini, DeepSeek, Qwen, GLM, Kimi | koda-base, koda-pro and the external models the Koda team has added |
| Where requests are processed | By the foreign model vendors via the RuAPI gateway (operated by CYBERBIT PTE. LTD., Singapore) | koda-base and koda-pro inside Russia under Federal Law 152-FZ, according to Koda |
| Where it works | Koda Desktop in Settings, Koda CLI via /model, the plugin via config.yaml | All Koda products, including unlimited autocomplete |
Based on kodacode.ru/tariffs and docs.kodacode.ru, checked 3 Nov 2026. It is not either-or: your model and Koda’s models can sit in the same list, and you can switch between them at any time.