header-langage
简体中文
繁體中文
English
Tiếng Việt
한국어
日本語
ภาษาไทย
Türkçe
Scan to Download the APP

OpenClaw has restored Claude CLI as the default backend within two days, after it was recently removed due to an Anthropic ban.

BlockBeats News, April 8, Open-source personal AI assistant OpenClaw released v2026.4.7. The most notable change is the restoration of Anthropic Claude CLI as the default local backend for new users, including onboarding flow, model authentication guidance, and Docker deployment paths. Just two days ago, version v2026.4.5 removed Claude CLI from the default backend, keeping only the compatible run with existing configurations. The attitude towards Claude CLI in the two versions has made a 180-degree turn. Background: Last week, Anthropic banned OpenClaw from using Claude subscription quotas, and the controversy caused by its abuse detection mechanism is still brewing.


Another signal in this version is the addition of Google Gemma 4 model support, including inference mode compatibility and thinking semantic adaptation. Gemma 4 has not been officially released, and OpenClaw's early access means that the model has entered the testing phase with external partners.


Other key updates:

1. Added openclaw infer unified inference command-line entry, integrating various inference tasks such as text, image, voice, embedding, and web search
2. Restored memory-wiki memory stack, providing structured knowledge statements, evidence fields, contradiction detection, and timeliness search
3. Media generation tool added cross-vendor automatic degradation, automatically switching to alternatives when preferred vendor authentication fails, and mapping parameters such as size, duration to the nearest options supported by the target vendor
4. Added pluggable compression engine, plugins can replace the built-in LLM digest pipeline
5. Added Arcee AI vendor and Ollama visual model auto-detection

Security fixes are prominent, involving CSRF protection to detach the request body to prevent credential leakage, block Java/Rust/Kubernetes environment variable override attacks, Feishu document upload path bypass, QQ Bot media download path injection, ClawHub plugin package SHA-256 verification, and more than ten others.

举报 Correction/Report
Correction/Report
Submit
Add Library
Visible to myself only
Public
Save
Choose Library
Add Library
Cancel
Finish