Free · Open source · OpenAI-compatible

Zero cost.

A free, OpenAI-compatible API that runs the DeepSeek and Qwen web clients, the official GigaChat API, the OpenCode Zen gateway and the unofficial Alice and Duck.ai endpoints server-side from your own provider tokens. No paid keys, no quotas.

Quick start Star on GitHub
Python 3.10+ Docker · GHCR
Features

Everything the paid APIs have.
None of the bills.

A drop-in replacement for the official OpenAI API - swap base_url and your existing client code just works.

OpenAI-compatible

GET /v1/models and POST /v1/chat/completions in the official format. Change base_url, keep your client.

Streaming

Live streaming with data: chunks and data: [DONE], plus thinking traces as reasoning_content.

Thinking traces

DeepSeek and Qwen reasoning exposed to your app - streamed live, on both providers.

Tool calling

Emulated tools, tool_choice and parallel_tool_calls with finish_reason: "tool_calls" responses.

JSON mode

response_format with json_object and json_schema - structured output for your agents.

Web search

Turn on grounded, up-to-date answers with the search flag on DeepSeek models.

File attachments

Images and text files as base64 or data URIs. Vision, OCR and file analysis per model.

Models

Six providers.
One OpenAI API.

Route by model name - deepseek-*, qwen*, GigaChat*, opencode/*, alice or a Duck.ai model. All optional, all can run together.

DeepSeek

  • deepseek-v4.1-flash thinking · search · files · vision

Read live from the web client settings, so only the model types the account is granted are listed. Web search via the search flag; reasoning via the thinking flag or reasoning_effort.

Qwen

Qwen

  • qwen3.8-max top model
  • qwen3.7-plus fast
  • ... fetched live from the account

Thinking and search built in. The model list is read from the account and refetched on a timer, so new models appear automatically.

G

GigaChat

  • GigaChat-2 lite · text
  • GigaChat-2-Pro tools · vision
  • GigaChat-2-Max tools · vision
  • GigaChat-3-Pro tools · vision

Official GigaChat API with a free freemium quota. Needs a Studio authorization key. The list is read from the account and refetched on a timer. Images work on Pro, Max and Ultra only.

Z

OpenCode Zen

  • opencode/gpt-5.6-sol tools · vision
  • opencode/claude-sonnet-5 tools · vision
  • opencode/kimi-k3 tools
  • ... fetched live from the gateway

The curated OpenCode gateway at opencode.ai/zen, metered per token. Needs an API key. Some ids collide with Qwen or DeepSeek, prefix those with opencode/. Only the chat completions half of the catalogue is served.

Я

Yandex Alice

  • alice no key needed
  • alice-ai no key needed
  • yagpt no key needed

Unofficial, opt-in via ALICE_ENABLED=1, off by default. Yandex has no public API here, so it uses an undocumented internal protocol that can break at any time. Stateless and no incremental text, so history is folded into one prompt. At least one message is required, tool calls are not supported, and max_tokens trims the answer.

D

Duck.ai

  • gpt-5.4-mini no key needed
  • gpt-5.6-luna no key needed
  • claude-haiku-4-5 no key needed
  • mistral-small-2603, tinfoil/gemma4-31b, tinfoil/gpt-oss-120b

Unofficial, opt-in via DUCKAI_ENABLED=1, off by default. DuckDuckGo has no public API here, so it uses an undocumented internal protocol and a browser fingerprint attestation that can break at any time. Needs Node.js on the host. Refuses datacenter addresses outright, so run it from a residential or mobile line.

M

Mistral Le Chat

  • mistral-small-latest
  • mistral-medium-latest
  • mistral-large-latest, magistral-medium-latest, codestral-latest, mistral-ocr-latest

Unofficial, opt-in via MISTRAL_ENABLED=1 plus MISTRAL_LOGINS with email:password pairs, off by default. Mistral has no public API here, so the provider speaks the Le Chat mobile flow with an account session. Free accounts are rate limited per message count. Can break at any time.

Quick start

Up and running in
under a minute.

One-command install on Windows, Linux and macOS. No Python setup gymnastics required.

All you need is a free DeepSeek or Qwen token, a GigaChat Studio authorization key, or an OpenCode Zen API key - the script does the rest.

PowerShell 
irm https://raw.githubusercontent.com/FANATFANATA/DanyAPI/prod/docs/install.ps1 | iex
Linux / macOS 
curl -fsSL https://raw.githubusercontent.com/FANATFANATA/DanyAPI/prod/docs/install.sh | bash

The script clones the repo, installs dependencies, creates .env, live-checks your provider tokens and tells you how to start the server. It even auto-updates itself on each start. Prefer not to dig tokens out of browser storage by hand? Run docs/token_utility.sh (or docs\token_utility.bat on Windows) and it pulls them out of your browser for you.

Docker · GHCR 
docker run -d -p 8000:8000 \
  -e DEEPSEEK_TOKENS="token1,token2" \
  -e QWEN_TOKENS="token3" \
  -e GIGACHAT_KEYS="<authorization_key>" \
  -e OPENCODE_KEYS="<zen_api_key>" \
  -e ALICE_ENABLED=1 \
  -e DUCKAI_ENABLED=1 \
  ghcr.io/fanatfanata/danyapi:latest

Prebuilt image, pushed on every push to prod and dev plus every version tag. The native PoW solver is compiled into the image for maximum speed.

FAQ

Questions?
Answered.

Is it really free?

Yes. DanyAPI uses the internal APIs of the free web clients chat.deepseek.com and chat.qwen.ai through accounts made from your own free provider tokens, plus the free tier of the official GigaChat API and the OpenCode Zen gateway. The unofficial Alice and Duck.ai providers need no credentials at all. No billing, no quotas.

Do my users need an API key?

When running DanyAPI locally or in private hosting, no API key is required (pass any dummy value).

Which providers and models?

Model lists are not hardcoded: DeepSeek (deepseek-v4.1-flash and whatever other types the account is granted), Qwen (qwen3.8-max, qwen3.7-plus, ...), GigaChat (GigaChat-2, GigaChat-2-Pro and the rest), OpenCode Zen (opencode/gpt-5.6-sol, opencode/kimi-k3, ...) and Duck.ai (the free tier models) are read from the provider endpoints and refetched on a timer, while Alice (alice, yagpt) serves its own aliases. Route by model name; all of them can run at once.

Are there rate limits or will my token get banned?

Requests go through the free web clients at a human-like pace. Add more tokens to the pool for extra parallelism - the project stays within normal usage, but treat free tokens as best-effort.

Is there usage tracking?

GET /v1/usage returns token usage totals, per-model and per-user breakdowns. Disable with DANYAPI_USAGE_ENABLED=0.

Free models.
Your API.

Install it now, or drop a star and follow the channel.

GitHub Telegram channel