Deepseek's POW(Proof of Work) Challange solver

A reimplimentation of deepseek's web chat pow solver with the help of gemini in rust and wasm to get near native speed in python using wasmtime.\

It is made for my project https://github.com/AmanCode22/deeperseeker which is the perfect docker supported and non-docker supported claude desktop app supported deepseek website reverse engineered project.

Building Yourself

The .wasm file is uploaded arleady in releases but if you wanna build it yourself then

• Clone the repo
• Install rust
• Add wasm32-unknown-unknown target( Run rustup target add wasm32-unknown-unknown)

Then run following command to build the wasm file

cargo build --target wasm32-unknown-unknown --release

Wasm file can be found in target/wasm32-unknown-unknown/release/deepseek_pow_solver.wasm

I made this as to solve pow challange in python when i tested took 66.3s with pure python implimentation for same challange and 0.3s with this implimentation. Deepseek's orignal wasm file was undocumented so that is why I reimplimented.

Usage Example

Install python and wasmtime(using pip install wasmtime) Use the following script to run this and add binary path in wasm_path variable

import wasmtime
wasm_path="ADD_YOUR_WASM_LIB_PATH"
engine = wasmtime.Engine()
with open(wasm_path, "rb") as f:
  wasm_bytes = f.read()
module = wasmtime.Module(engine, wasm_bytes)
store = wasmtime.Store(engine)
linker = wasmtime.Linker(engine)
instance = linker.instantiate(store, module)
memory = instance.exports(store)["memory"]
alloc_func = instance.exports(store)["alloc"]
solve_func = instance.exports(store)["solve_pow"]
def solve(challenge_hex, salt, expire_at, difficulty):
    def write_string(text):
        data = text.encode("utf-8")
        ptr = alloc_func(store, len(data))
        mem = memory.data_ptr(store)
        for i in range(len(data)):
            mem[ptr + i] = data[i]
        return ptr, len(data)

    ch_ptr, ch_len = write_string(challenge_hex)
    salt_ptr, salt_len = write_string(salt)

    result = solve_func(
        store, ch_ptr, ch_len, salt_ptr, salt_len, expire_at, difficulty
    )

    
    if result < 0:
        result = result + 0x10000000000000000

    return result if result != 0xFFFFFFFFFFFFFFFF else None

print("Answer:",solve("34d4c336676aa2e83c3308148e12ba8bc5e77ccb2fc12eeea1046e20c4c64eec","6c3a962c828dd81d7d69",1780227446451,144000))

Deepseek api returns json like below

{
    "code": 0,
    "msg": "",
    "data": {
        "biz_code": 0,
        "biz_msg": "",
        "biz_data": {
            "challenge": {
                "algorithm": "DeepSeekHashV1",
                "challenge": "34d4c336676aa2e83c3308148e12ba8bc5e77ccb2fc12eeea1046e20c4c64eec",
                "salt": "6c3a962c828dd81d7d69",
                "signature": "f68a3f7b13fde1a3c5f484869bed76d82160e39b401172865e31485219682b74",
                "difficulty": 144000,
                "expire_at": 1780227446451,
                "expire_after": 300000,
                "target_path": "/api/v0/chat/completion"
            }
        }
    }
}

And it expects a base64 string like

eyJhbGdvcml0aG0iOiJEZWVwU2Vla0hhc2hWMSIsImNoYWxsZW5nZSI6IjM0ZDRjMzM2Njc2YWEyZTgzYzMzMDgxNDhlMTJiYThiYzVlNzdjY2IyZmMxMmVlZWExMDQ2ZTIwYzRjNjRlZWMiLCJzYWx0IjoiNmMzYTk2MmM4MjhkZDgxZDdkNjkiLCJhbnN3ZXIiOjg2MDIyLCJzaWduYXR1cmUiOiJmNjhhM2Y3YjEzZmRlMWEzYzVmNDg0ODY5YmVkNzZkODIxNjBlMzliNDAxMTcyODY1ZTMxNDg1MjE5NjgyYjc0IiwidGFyZ2V0X3BhdGgiOiIvYXBpL3YwL2NoYXQvY29tcGxldGlvbiJ9

This base64 string decodes into a json

{
   "algorithm":"DeepSeekHashV1",
   "challenge":"34d4c336676aa2e83c3308148e12ba8bc5e77ccb2fc12eeea1046e20c4c64eec",
   "salt":"6c3a962c828dd81d7d69",
   "answer":86022,
   "signature":"f68a3f7b13fde1a3c5f484869bed76d82160e39b401172865e31485219682b74",
   "target_path":"/api/v0/chat/completion"
}

So, here we have to put the answer that we get from the script and other values that we got from sever itself. Then base64 encode it like shown above and can be used as x-ds-pow-response header.

Provenance & Legal Statement

This is an independent, clean-room reimplementation of the DeepSeekHashV1 proof-of-work algorithm, created to enable interoperability with the author's own DeepSeek account. No source code, binaries, or assets from DeepSeek's web application were copied, decompiled, or redistributed; the algorithm was reconstructed from observed input/output behavior analyzed with the assistance of AI. All code in this repository is original. The name "DeepSeek" is used solely to identify compatibility; this project is not affiliated with, endorsed by, or sponsored by DeepSeek.