cv-superding/dsh-deepseek-web-login
Unofficial DSH (DeepSeek Harness) plugin: use chat.deepseek.com web models as an LLM provider - browser-login capture, PoW solving, SSE streaming, prompting-based tool calls.
What it solves
This project allows you to use the free web version of DeepSeek (chat.deepseek.com) as a backend for DSH (DeepSeek Harness) agents. It bypasses the need for an official API key by capturing and reusing your browser's authenticated session, enabling you to run agents with DeepSeek's web-based features like thinking processes, image understanding, and tool calling without consuming paid API credits.
How it works
Instead of acting as a proxy or scraping the website, this tool acts as an unofficial client for DeepSeek's private web APIs. It captures your authentication headers and cookies once via a dedicated browser profile, then constructs its own requests directly to the DeepSeek backend. It handles complex tasks like solving Proof-of-Work (PoW) challenges using the browser's own WebAssembly code, managing SSE (Server-Sent Events) streams, and uploading files. It uses your system's Edge or Chrome browser to ensure the network fingerprint matches a real user, significantly reducing the risk of being flagged as an automated bot.
Who it’s for
This tool is for DSH users who want to leverage DeepSeek's advanced reasoning and multimodal capabilities for free, and who are comfortable managing browser-based authentication and occasional manual configuration to maintain access.
Highlights
- No API Key Required: Uses your existing web session to drive agent operations.
- Native-like Fingerprinting: Uses your system's Chrome or Edge browser to mimic real user traffic, avoiding common anti-bot triggers.
- Full Feature Support: Supports thinking streams, image uploads, and tool calling by parsing the web interface's private API protocols.
- Session Management: Automatically handles session persistence and cleanup to keep your chat history clean and avoid account-level rate limits.
- Configurable Throttling: Includes built-in request interval randomization and rate-limiting to mimic human interaction patterns.
Related
- Project
- Project
- Project
- Project