Changelog
New features, improvements, and fixes in Workshop.ai
- Sub-agent delegation Spin up specialized sub-agents to handle complex multi-step tasks.
- Parallel tool execution Read-only tools are now executed in parallel, making multi-file operations faster.
- Google Gemini 3.1 Pro and Flash Google Gemini 3.1 Pro and Flash models are now available.
- Z.ai GLM-5.1 Z.ai GLM 5.1 frontier class models are now available.
- OpenAI GPT-5.4 OpenAI GPT 5.4 models are now available.
- Redesigned planning mode Planning mode has been redesigned with a clearer flow for reviewing plans before the agent starts building.
- Local models use HuggingFace Hub cache Local models now use your existing HuggingFace Hub cache, so you won't re-download models you already have.
- General bug fixes and stability improvements.
- Sub-agent delegation The agent can now break complex tasks into subtasks and delegate them to specialized sub-agents that work independently. This lets the agent tackle larger, multi-step projects — each sub-agent focuses on its piece of the problem and reports back with results.
- Z.ai GLM-5.1 model support Z.ai GLM-5.1 is now available as a model choice via OpenRouter.
- OpenAI GPT-5.4 support Added support for OpenAI GPT-5.4 models.
- General bug fixes and UX improvements.
- Parallel tool execution The agent can now run multiple tools simultaneously, with real-time progress indicators showing what's happening in parallel — making complex tasks significantly faster.
- Google Gemini 3.1 support Google Gemini 3.1 is now available, with full support for extended thinking and multi-turn conversations.
- Planning mode redesign Planning mode has a cleaner interface, smarter plan generation, and tighter integration with the agent's task flow.
- Subagent delegation The agent can now spawn and coordinate specialized subagents to tackle different parts of a task in parallel, enabling more sophisticated multi-step projects.
- Auto-enable connectors Connectors are now automatically activated when the agent detects they're needed — no more manual setup before building.
- Faster deployments Deployed apps now build significantly faster thanks to improved caching.
- Faster project startup and general performance improvements.
- Optimized subagent performance for better quality at lower cost.
- Large image uploads no longer fail mid-conversation.
- Deployed apps that crash now show a proper error page instead of silently hanging.
- General UX improvements and robustness fixes.
- Google Gemma 4 local models Run Google Gemma 4 models locally — new models appear automatically in the local models tab, no app update needed.
- AMD APU unified memory detection Workshop now correctly detects AMD APU unified memory, giving you more accurate local model recommendations on AMD-powered laptops and desktops.
- Workshop now uses your OS trust store for SSL verification, fixing connection failures behind enterprise proxies with custom SSL certificates.
- Fixed a rare crash where certain message formats could cause the agent to hit token limits unexpectedly.
- General stability improvements and bug fixes.