Version 3.8.28
π Changelogβ
π v3.8.28β
π
October 11, 2026
Featuresβ
- Added Settings βΊ Automation βΊ AI Agent Runs, where you can review every AI agent run with its harness, status, model, tool calls against the budget, token usage and duration, and filter by harness, status and date. Opening a run shows each step, including model and tool calls, fallback switches, intermediate messages and retries, along with its instructions, input and result.
- Function nodes can now have their own settings and outputs. When a custom function describes its parameters and outputs, the Function node shows a description and a field for each parameter, and its outputs are added automatically. Custom function editors include a JSON editor for this definition.
Enhancementsβ
- Updating an installed app no longer breaks the AI tools and Function nodes in your own flows and AI nodes that use the app's functions.
- AI agent answers, intermediate messages and notes longer than a channel allows (for example, 4,096 characters on WhatsApp) are now split into several messages. A message that still cannot be sent no longer leaves the flow waiting.
- In AI Harnesses, you can now create harnesses that use your own credentials, while MonoChat's built-in harnesses remain available to every workspace. The AI Agent node now shows readable labels in its selections, and the Models section asks you to choose a harness first.
Bug Fixesβ
- Fixed AI agent runs failing with "Cannot read properties of undefined" when a step only called tools.