WayToClawEarn
Medium impactHacker News + GitHub

DeepClaude community hot discussion: Behind the 17 times saving, data privacy becomes the biggest concern

The DeepClaude open source project received 544 votes and 230 discussions on Hacker News in 15 hours. Community testing confirmed that "no difference can be felt at all", but the DeepSeek API does not allow users to opt out of training data collection, causing widespread concern.

Edisen Lu · WayToClawEarnVia Hacker News + GitHubPublished May 4, 2026Updated Sep 28, 2026

Reviewed from public sources · AI-assisted drafting under editorial oversight. How we work · Original source

Public-source compilation

Synthesized from public posts/docs. Prefer the original source for primary claims.

How we review content · Primary source · Hacker News + GitHub

Core conclusion

The DeepClaude open source project quickly ignited the developer community after its release - it received 544 votes and 230 discussions on Hacker News in 15 hours, and received 726 stars on GitHub. This project allows developers to switch Claude Code's autonomous Agent loop backend to DeepSeek V4 Pro, and the API cost drops from Anthropic's $15/M output token to DeepSeek's $0.87/M, a drop of 17 times.

But a key controversy emerged in community discussions: DeepSeek's API terms do not allow users to opt out of model training data collection. For developers working on non-confidential projects, "there is no difference at all", but when it comes to sensitive corporate code, data flow becomes a choice that must be faced.

Key Points

  • Time of incident: 2026-05-03 to 2026-05-04 -Affected objects: developers and teams using Claude Code, especially those concerned about API costs and data privacy
  • Core change: Low-cost AI Coding toolchain emerges, but comes with data privacy trade-offs

DeepClaude technology principle

The core idea of DeepClaude is very simple: Claude Code CLI determines the API call target through the environment variables ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN. DeepClaude points these two variables to DeepSeek's Anthropic-compatible interface, so that the complete agent cycle of Claude Code - file reading and writing, Bash execution, Agent generation, sub-Agent calling - all remain unchanged, only the "thinking brain" is replaced by DeepSeek.

The project provides a command to switch between different backends:

terminal
deepclaude # DeepSeek V4 Pro($0.87/M )
deepclaude --backend or # OpenRouter (,US )
deepclaude --backend fw # Fireworks AI (,$3.48/M )
deepclaude --backend anthropic # Anthropic($15/M ,)

rapind HN 「,。」 aftbit —— 5 Shell

terminal
export ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic
export ANTHROPIC_AUTH_TOKEN=sk-your-key
export ANTHROPIC_MODEL=deepseek-v4-flash
export CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1
exec claude $@

Anthropic MaxDeepClaude (DeepSeek)
(10/)$200/~$20/90%
(25/)$200/~$50/75%
( Token)$200/~$80/60%

。 DeepSeek V4 Pro LiveCodeBench 96.4% ,「」。

AI coding cost comparison chart

,。DeepSeek API 。 AI Coding Agent ,。

lhl HN 「OpenRouter zdr: true() data_collection: deny,。HuggingFace Inference DeepSeek V4 Kimi K2.6, Together AI、DeepInfra ,。」

,

  1. ** OpenRouter **——(zdr)
  2. ** HuggingFace Inference**——
  3. ****—— DeepClaude
  4. ** Anthropic**——--backend anthropic

DeepClaude --switch ,。 DeepSeek(), Anthropic(), Fireworks AI()。

「,」, AI 。

DeepSeek、OpenAI、ChatGPT、Claude、Claude Code、OpenRouter、HuggingFace、Fireworks AI、Anthropic、GitHub.

Internal link guidance

anthropicdeepseeksecuritycoding

View source →

Educational reference only: cases summarize public sources and may use AI-assisted drafting under editorial review. Not financial advice; outcomes are not guaranteed.