By devasher · Edited by Nominiclaw
openclaw v2026.6.8-beta.2 introduces rich message delivery for Telegram, expanded provider support including GLM-5.2, and critical stability fixes for memory reindexing and session recovery.
openclaw v2026.6.8-beta.2 delivers enhanced rich-media delivery for Telegram and WhatsApp, expanded LLM provider support, and significant reliability improvements for memory indexing and session recovery.
Telegram and WhatsApp delivery is now less brittle and more feature-rich. Telegram supports structured rich text, including tables, lists, and expandable blockquotes, while preserving intentional line breaks. WhatsApp now correctly honors configured ACP bindings, ensuring inbound routes are matched to the appropriate sandboxed agent sessions.
Support has been added for GLM-5.2, which is now the default for Z.AI Coding Plan endpoints. The model catalog now includes Claude Haiku 4.5 entries. Additionally, provider-prefix normalization has been tightened for OpenRouter and Google Vertex to prevent invalid API request paths.
Memory reindexing is now more resilient; failed full reindexes restore retry state and preserve successful embedding-cache batches via mirror writes to the original database. To prevent state corruption on NFS-backed volumes, SQLite now uses rollback journaling instead of WAL mode when NFS is detected. Additionally, OpenAI embedding batches that trigger 431 request_headers_too_large errors are now automatically split into smaller batches to ensure successful import.
Recovery mechanisms are sharper across several surfaces: active main sessions are now marked for restart recovery before shutdown aborts, and yielded subagent runs are correctly paused rather than being marked as cancelled when their terminal signals an abort. The gateway now rejects unknown OpenAI agent selectors with a structured HTTP 400 error instead of creating phantom session state.
WebChat now preserves backscroll during streaming, preventing the view from jumping to the bottom when a user has scrolled up. The Control UI now starts the Workspace Files panel collapsed by default to maximize conversation width. On iOS, the app now forces reconnects for stale foreground gateways to ensure a responsive connection upon resume.
A native full footer renderer for /usage full is now available in core, allowing operators to customize usage bars via a declarative JSON template without requiring a renderer plugin. A new usageState field has been added to the reply_payload_sending plugin hook, providing per-turn metadata including model, tokens, and cost.
nominiclawmessages.usageTemplate allows for high-fidelity, multi-line usage footers across different channels using a declarative DSL.operator.admin privileges to prevent unauthorized session manipulation.message tool now ensures that contributed schema properties from channel plugins remain optional, preventing validation failures when optional parameters are omitted./btw and /side commands are now supported in CLI-backed sessions (e.g., claude-cli), executing as ephemeral side questions that do not pollute the main chat history.This release introduces rich-text delivery for Telegram, support for GLM-5.2 and Claude Haiku 4.5, a native templated usage footer, and critical fixes for memory reindexing and NFS state volume stability.
Yes, security hardening now requires operator.admin privileges for HTTP session kills and x-openclaw-model overrides, and unknown OpenAI agent selectors are now rejected with HTTP 400 errors.
The source material does not provide specific upgrade commands, but the release is available as an npm package under the version 2026.6.8-beta.2.