Documentation

Everything about the rbug provider API — endpoints, limits, errors, pricing and SDK setup. No hidden details: every cap and behavior is documented below.

Overview

rbug provider is an AI gateway behind one API key. It exposes OpenAI-compatible and Anthropic-compatible endpoints and routes requests to a catalogue of 40+ models, with 500,000 free output tokens on every account and pay-as-you-go billing after that. Point any OpenAI or Anthropic SDK at https://rbug.tech/api/v1 and it works.

• Streaming (SSE), system prompts, tool calling and reasoning fields are supported (model-dependent).
• An API key + a base_url swap is all most apps need.
• A built-in chat playground lives in the console.

Quickstart

1. Sign in with Google or email.
2. Open Console → API keys and create a key (starts with rbp-).
3. Make your first request:

# curl
curl https://rbug.tech/api/v1/chat/completions \
  -H "Authorization: Bearer rbp-your-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4.1-flash",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

# Python (OpenAI SDK)
from openai import OpenAI

client = OpenAI(base_url="https://rbug.tech/api/v1", api_key="rbp-your-key")

reply = client.chat.completions.create(
    model="deepseek-v4.1-flash",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(reply.choices[0].message.content)

// Node.js (openai package)
const client = new OpenAI({ baseURL: "https://rbug.tech/api/v1", apiKey: "rbp-your-key" });
const reply = await client.chat.completions.create({ model: "deepseek-v4.1-flash", messages: [{ role: "user", content: "Hello!" }] });

Authentication

Send your key either way — both are accepted:

Authorization: Bearer rbp-your-key
# or
x-api-key: rbp-your-key

• Keys are created, rotated and revoked in Console → API keys. Only a hash is stored server-side; the full key is shown once at creation.
• GET /v1/models and GET /v1/models/{id} are public — no key needed, so SDKs can discover models.
• All chat endpoints require a valid key. Invalid keys return 401 with invalid_api_key.

List models

GET/v1/modelspublic

Returns an OpenAI-style list object. 40+ models at the time of writing (upstreams rotate models, so always check live).

curl https://rbug.tech/api/v1/models

{
  "object": "list",
  "data": [
    { "id": "deepseek-v4.1-flash", "object": "model", "owned_by": "workbuddy", "free": true },
    ...
  ]
}

GET/v1/models/{id}public

Returns a single model object, or 404 model_not_found.

Chat completions

POST/v1/chat/completions

Standard OpenAI chat completions schema. Supported fields:

Non-streaming responses are complete chat.completion objects with a usage block (prompt_tokens, completion_tokens, total_tokens) and a unique id.

# Streaming
curl https://rbug.tech/api/v1/chat/completions \
  -H "Authorization: Bearer rbp-your-key" \
  -H "Content-Type: application/json" \
  -d '{"model":"deepseek-v4.1-flash","stream":true,"messages":[{"role":"user","content":"Hello"}]}'

data: {"object":"chat.completion.chunk","model":"deepseek-v4.1-flash","choices":[{"index":0,"delta":{"content":"Hi"}}]}
...
data: [DONE]

Legacy completions

POST/v1/completions

For older tools and SDKs that still use the text-completions API (SillyTavern, legacy libraries). The prompt is converted into a chat message and the reply is returned in the classic text_completion shape — streaming included (data: chunks + [DONE]).

A base-URL health check is also available: GET /v1 returns a small info object listing all endpoints.

Anthropic messages

POST/v1/messages

Anthropic-compatible endpoint for the Anthropic SDK and tools like Claude Code. Supports system, max_tokens, stream and returns Anthropic-shaped responses (type: "message", content blocks, usage.input_tokens/output_tokens).

export ANTHROPIC_BASE_URL=https://rbug.tech/api/v1
export ANTHROPIC_API_KEY=rbp-your-key
# any Anthropic SDK or Claude-compatible tool now routes through rbug
# (both https://rbug.tech/api and https://rbug.tech/api/v1 work - the gateway
#  normalizes the duplicated /v1 that SDKs append)

Verified with the official Anthropic SDK: messages.create, streaming (messages.stream) and system prompts all work.

Count tokens

POST/v1/messages/count_tokens

Returns an estimated token count for the request:

{ "input_tokens": 7 }

Estimation is character-based (~4 characters per token). Treat it as an approximation, not an exact tokenizer.

Limits

Every limit, stated plainly:

• Global concurrency: the server serves up to 25 chats at once across all users; beyond that you get 503 server_busy — retry in a few seconds.
• Input size: API requests are forwarded without trimming — the only limits are the 32MB request body and the model's own context window. The console chat sends only the newest ~72,000 characters of conversation.
• Context window: GET /v1/models reports a context_length for every model — at least 1,000,000 tokens, and larger where the model truly supports it (for example auto-free at 2,000,000). Because requests are forwarded untrimmed, the practical ceiling is the upstream model's own window; a request past it comes back as an upstream_error.
• Output size: max_tokens is clamped to 32–4096.
• Best-effort upstreams: the free models run on shared capacity. Occasional 502, 504 or 429 under load is normal — retry or switch model.

Errors

Errors use the OpenAI error shape: { "error": { "message": "...", "type": "...", "code": "..." } }.

Pricing & credits

• Every account includes 500,000 free output tokens — no card, no signup friction.
• After the free allowance, usage is billed from your credit balance at $0.165 per 1M input tokens and $0.66 per 1M output tokens. Fractions of a cent are accumulated and whole cents are deducted as you go.
• With a positive balance you also get the raised rate limits shown above. When the balance reaches zero, requests return 402 free_limit_reached until you top up.
• Top-ups are crypto-only: USDT (Tron network), BTC, LTC, ETH — shown on the Credits page with the exact addresses. Orders are reviewed and credited manually after confirmation.
• No subscriptions, no auto-billing, no expiration on credit.

Console & chat

The console contains:

• Chat — a full ChatGPT-style playground: multiple conversations with search and rename, edit & resend, regeneration history (◀ 2/3 ▶), thumbs feedback, export as Markdown, voice input (where the browser supports it), keyboard shortcuts (Ctrl/Cmd+K new chat, / focus input, Esc stop). The + menu enables look-up tools: web search, Wikipedia and GitHub — these only fetch public pages (600 KB text cap, no code execution).
• API keys — create, list and revoke rbp- keys.
• Credits — deposit addresses and payment status.
• Usage — output-token counter against your allowance.
• Orders — credit top-up history.
• Support — contact the team.

SDK setup

OpenAI SDKs

# Python
client = OpenAI(base_url="https://rbug.tech/api/v1", api_key="rbp-your-key")

// JavaScript
const client = new OpenAI({ baseURL: "https://rbug.tech/api/v1", apiKey: "rbp-your-key" });

Anthropic SDK / Claude Code

export ANTHROPIC_BASE_URL=https://rbug.tech/api/v1
export ANTHROPIC_API_KEY=rbp-your-key

opencode (CLI)

Add to ~/.config/opencode/opencode.jsonc (set model to a tool-capable model — opencode needs function calling):

{
  "$schema": "https://opencode.ai/config.json",
  "model": "rbug/deepseek-v4.1-flash",
  "provider": {
    "rbug": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "rbug provider",
      "options": { "baseURL": "https://rbug.tech/api/v1" },
      "models": {
        "deepseek-v4.1-flash": { "name": "DeepSeek V4.1 Flash" },
        "north-mini-code": { "name": "North Mini Code" },
        "GLM-5.3-Flash": { "name": "GLM-5.3 Flash" }
      }
    }
  }
}

Then run opencode auth login → Other → provider id rbug → paste your key, and restart opencode. Verified with the same ai SDK opencode uses: generateText, streamText and multi-step tool calls all pass.

Data & privacy

• Message content is not stored server-side. API requests are proxied in memory; the console chat is saved in your browser's local storage.
• We store: your account (name/email from sign-in), hashed API keys, and per-account usage counters (token totals) used for the free-tier limit.
• Requests are forwarded to the model hosts that serve each model, under their own policies.
• Keys are shown once at creation and stored only as SHA-256 hashes.

FAQ