OpenClaw
OpenClaw 2026.8.1-beta.2 tightens secret egress and model runtime switching
OpenClaw's latest beta binds shared-store secrets to exact HTTPS hosts and adds GPT-5.6 Ultra support across Sol, Terra, and Luna runtimes.
read --wire →
$ tail -f topics/ai.log
Daily updates from the AI and machine-learning ecosystem - PyTorch, Hugging Face, the Anthropic and OpenAI SDKs, local LLMs, and the agent frameworks.
256 updates across 10 projects this week.
One calm review of what shipped across AI / ML - the commits, releases, and security advisories that matter. Every Monday, with security advisories same-day. Free, unsubscribe in one click.
We'll start you on the top AI / ML projects - refine anytime. · Read a sample issue →
OpenClaw
OpenClaw's latest beta binds shared-store secrets to exact HTTPS hosts and adds GPT-5.6 Ultra support across Sol, Terra, and Luna runtimes.
read --wire →
OpenClaw
Control UI terminals and model setup now work in explicit-agent fleets after a fix for a race condition that surfaced during Gateway restarts.
read --wire →
OpenClaw
The Control UI now uses one channel picker and one model picker across every surface, fixing both visual divergence and a data corruption bug from sentinel mishandling.
read --wire →
OpenClaw
Clawsweeper's performance and compaction fixes for OpenClaw's model resolution are now shipping, with Slack workspace policy scoping for enterprise grids.
read --wire →
OpenClaw
ClawSweeper now routes dashboard GitHub App credentials through a single canonical stack, duplicating security-critical signing code no longer, while retiring a telemetry pipeline that had no production producer.
read --wire →
OpenClaw
OpenClaw unified secret-redaction and SSRF policy ownership to close a stale-redaction gap, while Clawsweeper throttled exact-review admission and deduped duplicate receipts to stop dead-lettering finished reviews at 20-30 per hour.
read --wire →
OpenClaw
Clawsweeper now acknowledges pull requests the moment their webhook arrives, closing a production gap where contributors waited minutes for any sign their PR was seen.
read --wire →
OpenClaw
openclaw mcp login now waits on the loopback port it registers, so browser approval completes instead of returning connection-refused.
read --wire →
OpenClaw
Critical fixes prevent process crashes on stream errors and stop forged proof headings from corrupting PR verdicts.
read --wire →
OpenClaw
ClawSweeper isolates public reads from privileged quota, and OpenClaw removes bodyless 400/413 patterns that caused false context compaction.
read --wire →
OpenClaw
OpenClaw fixed a zero-work drain loop in the auto-reply gateway and plugged QA lease leaks, both caused by recent policy changes.
read --wire →
OpenClaw
Crabbox fixes stuck Azure leases and untracked cloud resources after interrupted provisioning, while OpenClaw adds primary QA proof for core agent runtime behaviors.
read --wire →
OpenClaw
OpenClaw fixed a bug where delayed Control UI chat commands, including a queued /reset, could outlive the gateway connection and apply to a replacement session.
read --wire →
OpenClaw
OpenClaw's daily patches cut 500 lines of duplicated Slack test fixtures, consolidate sibling providers, and eliminate UI contract drift.
read --wire →
OpenClaw
OpenClaw's dev-channel updater now reuses the build cache across preflight and final checkout, cutting duplicate TypeScript declaration and Control UI builds.
read --wire →
OpenClaw
Four bug fixes shipped in openclaw/openclaw, including a memory safety fix for oversized local attachments and a Telegram login code convenience improvement.
read --wire →
OpenClaw
OpenClaw shipped fixes that stop delegated messages from routing to the wrong account and make tool schemas compatible with llama.cpp providers.
read --wire →
OpenClaw
One credential failure no longer disables message actions for healthy accounts.
read --wire →
OpenClaw
OpenClaw fixes a bug where installed external web-search providers went undetected on fresh installs, breaking env-var auto-detect.
read --wire →
OpenClaw
ClawSweeper removed the Claude CLI layer, added a built-in Cerebras provider, and shipped a deny-by-default environment allowlist for OpenClaw subprocesses.
read --wire →