Open WebUI v0.13.1: Coding Agents in the Model Picker

• Coding agents appear in the model picker on an Open Terminal that offers them. Each conversation uses one agent.
• Tether is a new desktop companion app for macOS, Windows and Linux.
• Auto is a third tool permission mode. It is off by default and administrators turn it on.
• Files whose extracted text is shorter than 50,000 characters are included in full by default.
• Preload files is off by default with native function calling.
• RAG_TOP_K and RAG_TOP_K_RERANKER now default to 10, and CHUNK_SIZE defaults to 1024.
• The release includes database schema changes and security fixes.

Coding Agents in the Model Picker Work Inside Chat Limits #

Coding agents now appear in the model picker when an Open Terminal offers them. The Open WebUI v0.13.1 release notes list Claude Code, Codex and Pi, and say you can chat with them or call them by voice, choose their reasoning effort and service tier, and approve their tool requests in the chat. Tool approval therefore happens in the same chat as the conversation, which suggests the approval flow is the place to review what an agent does on your terminal.

Agent chats have limits. Each conversation uses one agent, and agent chats cannot be regenerated, continued or edited and resent. If an agent's terminal goes away, its chats and new chats fall back to the default model, while earlier replies keep the agent's name. Autocompletion in agent chats uses the task model, or is skipped when there is no task model, so check that a task model is set if you expect suggestions.

Tether Connects Your Computer to Open WebUI Without Open Ports #

Tether is the new desktop companion app that ties your own computer to Open WebUI. After you sign in once, the computer appears under Your devices in the terminal menu, so models, coding agents and automations can work in your real files, repositories and tools. The release notes state that Tether connects out to Open WebUI, which is why no inbound ports need to be opened.

"Tether connects out to Open WebUI, so there are no ports to open"

Open WebUI v0.13.1 release notes

Its session is renewable and expires after 90 days without use by default. The lifetime is set with SESSION_EXPIRES_IN. Tether also resumes terminal output after a dropped connection and delivers desktop notifications when no Open WebUI page is open. Earlier desktop releases matter here: the v0.13.0 release notes say Open WebUI Desktop no longer switches between remote servers, and that existing remote-connection users should review Open WebUI Tether.

Auto Tool Permissions Send Tool Calls to a Scoring API #

Auto is off by default. Administrators can turn it on as a third tool permission mode next to Full access and Ask for approval. A review model checks each tool call and runs it immediately only when it judges that no confirmation is needed. The v0.13.1 notes state what the review sends: the tool, its arguments and your last few messages, to the configured scoring API.

Calls the review cannot assess, because of an error, a timeout or too much context, ask for approval instead of running. Before enabling Auto, check which scoring API is configured, since this suggests that tool arguments and recent messages leave Open WebUI for that service whenever the review runs.

Small Files Are Read in Full Instead of Retrieved #

Files under 50,000 extracted characters are now included in full. The threshold is set with RAG_FULL_TEXT_CHAR_THRESHOLD or in the admin Documents settings. The inclusion applies both when Preload files adds the content to the conversation and when the model reads an attached file with its file search tool, and the file viewer shows when it happens. Files at or above the threshold keep going through retrieval.

The Full Context Mode setting in the admin Documents settings is renamed Always Use Full Text in this release, and it includes complete documents regardless of the full text threshold. Administrators who relied on the old name should look for the new one before deciding how large attachments should be handled.

Attached Files and Tool Chats Change Shape in v0.13.1 #

Preload files is off by default when native function calling is used. Attached files are no longer added to the conversation automatically, and models read them with their file tools instead. Models without tool support, or without the Files built-in tool, cannot see attached files until Preload files is turned on in the model's Capabilities. Legacy function calling, and API requests that do not come from a chat in the web UI, keep the old behaviour.

The same release moves memory recall, the skills list and attached knowledge out of the system prompt and into hidden tool results after your message. The release notes connect this to cost: the start of the conversation stays the same between requests, so providers can serve it from their prompt cache. Custom pipes, however, now see a tool message as the last message, so pipes that assume a different layout need review.

Retrieval Defaults Grow Prompts for Existing Knowledge Bases #

Unless you saved your own values, RAG_TOP_K and RAG_TOP_K_RERANKER now default to 10 instead of 3, and CHUNK_SIZE to 1024 instead of 1000. Each query retrieves more text, so prompts with knowledge get longer. The new chunk size applies to files processed from now on. Existing knowledge keeps its chunks until it is reindexed, so a knowledge base can mix chunk sizes until that happens.

Upgrade Checklist for Open WebUI v0.13.1 #

1. Back up your database and all associated data before upgrading production, as the release notes recommend.
2. If you run a multi-worker, multi-server or load-balanced deployment, update all instances simultaneously. Rolling updates are not supported and will cause application failures due to schema incompatibility.
3. Check whether models without tool support need attached files, and turn on Preload files in their Capabilities if they do.
4. Check whether your retrieval values were saved in the admin settings, because only unsaved values move to the new defaults.
5. If you turn on Auto, confirm which scoring API receives tool calls and recent messages.
6. Update soon: the release contains security fixes. The notes say these fixes are not listed in the changelog, and that advisories will follow CVE assignment.

Also in This Release #

The v0.13.1 release notes list many more items, including tab keyboard shortcuts, split panes, Git tools and spreadsheet editing. The full list is in the release notes. Items that affect operations include:

• Tool calls that fail twice in a row with the same arguments now end the tool loop.
• OpenAI image generation defaults to GPT-IMAGE 2, and the shut-down DALL·E models are no longer offered.
• Terminal connections set to Session authentication accept an API key again.
• Redis Cluster deployments no longer fail chat sends while the personal terminal relay starts.
• Milvus knowledge base uploads no longer store chunks twice.
• The Tool Call Labels setting is off by default and is controlled with ENABLE_TOOL_CALL_LABELS.