Open WebUI 0.13.0 adds browser tabs and new agent controls
Open WebUI 0.13.0 brings browser and desktop tabs into its chat workspace, alongside files, terminals and previews. The official November 3 release also adds delegation across different models. For existing deployments, the immediate job is planning the database migration and reviewing agent limits before enabling longer workflows.
Browser and desktop access need a connected runtime
The browser tab can share a Chrome session between the user and model. It requires browser support on the selected Open Terminal and Chrome availability for automation. Desktop control similarly needs a terminal with desktop support. These are connections to an execution environment, not capabilities that appear simply because the chat interface was upgraded.
Open Terminal provides the files, shell and running processes behind the conversation. The connection guide recommends adding it through Settings > Admin > Integrations, in the dedicated Open Terminal section, then selecting it in chat. Native function calling is the recommended model setting. An ordinary external tool-server connection does not provide the same terminal sidebar.
The runtime documentation cautions that emitting a valid tool call is not enough for reliable multi-step work. A model must inspect output, recover from errors and verify results. Before relying on it, try a small representative task with disposable files and check the actual output.
Different models can take different jobs
The model editor lets an administrator or authorized editor configure which models a parent can delegate to. A user can receive only models they already have permission to access. Leaving the selection empty keeps delegated work on the current model. That makes separate research, implementation and review roles possible without granting access to an otherwise unavailable model.
The configuration reference lists sub-agents as disabled by default. Once enabled, foreground concurrency, background concurrency, tool-loop and returned-output caps default to unlimited. Concurrency limits are counted per worker process, so a setting is not automatically a fleet-wide ceiling. Sub-agent tool-loop limits are separate from the main response limit.
A normal response using native function calling now defaults to 1,024 tool-calling rounds, up from 256. A round can contain several calls. Setting CHAT_RESPONSE_MAX_TOOL_CALL_ITERATIONS to 256 retains the earlier ceiling. Longer runs can consume more compute or paid model calls, so define a budget and stopping point before treating an unlimited default as a production policy.
MCP Apps add controls inside a conversation
MCP Apps let compatible tools return an interactive interface inside chat. The documentation describes an isolated frame with declared network permissions, and each tool call initiated by the app requires confirmation. Native MCP connections use Streamable HTTP and are configured by administrators. App support does not make every MCP server an interactive app or make the selected model reliable at using its tools.
Back up and coordinate the upgrade
Version 0.13.0 changes the database schema. The release notes warn against rolling upgrades across mixed versions and recommend backing up the database and associated data. They also report security fixes without describing the vulnerabilities. No exploit or CVE detail should be inferred from that notice.
The official update guide recommends a pinned version and staging checks for production systems. For multiple replicas, stop the old instances, let one upgraded instance perform the migration, then return the others on the new version. Assigning migration responsibility to one instance does not make a mixed-version deployment compatible.
The separately installed Open WebUI Desktop app also has a changed role in its 0.1.x releases. It no longer switches between remote servers or bundles Open Terminal. Existing desktop users should check those changes before assuming their previous connection workflow remains available.
What is free, and what still costs money
Open WebUI remains free to use as-is. Its license includes branding requirements, so current versions should not be described as unrestricted BSD software. The license provides specific branding exceptions, including deployments with 50 or fewer end users in a rolling 30-day period. That threshold is a branding condition, not a general cap on free usage.
Terminals, the separate fleet orchestrator, requires an enterprise license for production. The enterprise page directs organizations to sales rather than publishing a standard price. Hosting, local inference hardware and any paid model APIs remain separate operating costs.
The practical choice is whether you want to operate this environment yourself. ByteForward’s Antigravity Remote Agents coverage explains a managed alternative and its access restrictions. For Open WebUI, start with a backed-up installation, a restricted workspace and one workflow you can verify. This report is based on release notes and documentation, not independent performance testing.
