BridgesLLM Portal v3.25.4
BridgesLLM Portal v3.25.4
Highlights
- Installer and updater flows now auto-apply the validated OpenClaw relay + Gemini compatibility patch set when needed instead of burying that repair behind a manual Settings step.
- Agent Chat and project chat are more truthful during live OpenClaw runs: capable runtimes now use real interrupt-and-steer behavior, running turns surface an immediate Thinking state, async follow-ups can finish cleanly even if the browser disconnects, and hidden Portal Backend RPC / heartbeat artifact text is stripped out of restored history.
- Project chat now runs in a saner repo context: project agents bind the repo into /workspace/project, assistant auto-commit is back after successful runs, transient .agent-* scratch files are shelved out of git operations, and revert/auto-commit flows stop confusing portal-maintenance state with user work.
- AI setup and provider status got a big honesty pass: Claude setup-token finishing no longer looks frozen, Codex/Gemini native CLI auth is sturdier, provider model IDs/fallbacks are normalized consistently, and provider cleanup removes stale auth/model config together.
- Windows / WSL beta messaging now tells the truth: local installs are labeled as experimental localhost test drives, while public hosting, stable share links, and custom-domain HTTPS remain VPS-first.
Also included
- Expanded compatibility status surfaces in Settings and Agent Chat, including relay + Gemini markers and clearer fallback wording.
- OpenClaw compaction-notice toggle in Settings.
- Better tool alias presentation for current OpenClaw/OpenCode-style tool names.
- Focused regression tests for AI-setup model normalization and OpenClaw CLI JSON parsing.
- Bundled Windows / WSL beta notes.