release aidev PRERELEASE SIG 3/5

OpenClaw 2026.6.8-beta.2: GLM-5.2 and Claude Haiku 4.5 catalog, web search opt-in change

Beta adding GLM-5.2 support and Claude Haiku 4.5 catalog entries with normalized provider-qualified model IDs, plus rich Telegram message rendering and WhatsApp ACP bindings. Behavior change worth checking: key-free web search providers (Parallel Free, DuckDuckGo, Ollama, Codex Hosted Search) are now explicit opt-ins and are no longer selected automatically when no API-backed provider is configured.

PUBLISHED2026-06-16
OBSERVED2026-08-11
AGE1mo
SOURCES1
  • Models: GLM-5.2 support and Claude Haiku 4.5 catalog entries; provider-qualified model IDs normalized across OpenRouter and Google Vertex paths; managed SecretRef auth; bounded model browsing; safer OpenAI/Anthropic tool-schema recovery
  • Web search behavior change: key-free providers (Parallel Free, DuckDuckGo, Ollama, Codex Hosted Search) are explicit opt-ins, not automatic fallbacks when no API-backed provider is configured
  • Channels: Telegram renders tables, lists, expandable blockquotes, preserved intentional line breaks and CLI-backed replies; WhatsApp honors configured ACP bindings
  • Usage footers: native full footer renderer for /usage and reply payload hooks, default template, fixed-decimal formatting, credential-aware limits, warnings for broken templates instead of silent bad output
  • Agent runtime: account-scoped DM send policy, generated media completions, auto-reply message-tool final replies, reset archive fallback reads, restart shutdown aborts, yielded subagent pauses, session identity in runtime prompts
  • Gemini CLI runs in an isolated runtime home using the selected OpenClaw OAuth/API-key profile, so ambient Google machine credentials cannot override it
  • Discord: generated auto-thread titles get a 60-second timeout and 4,096-token reasoning output budget, clamped to the model output cap
  • Memory/state: oversized OpenAI embedding batches split before 431s, QMD search stays available in transient mode, SQLite avoids WAL on NFS, full reindexes preserve rollback/cache recovery
  • Security: admin privileges required for HTTP session/model override surfaces; unknown OpenAI agent selectors rejected
  • Agent commands: /btw supported in CLI-backed sessions; CLI usage-error exits classified as failures, not successful runs
model-supportfeaturefixsecurity

COMMUNITY

No curated reactions recorded for this event. Facts and takes are kept in separate layers — community context is added by hand, never blended into the record above.