Skip to content
Projects

Skyzi000/open-webui-extensions

View on GitHub

Skyzi000/open-webui-extensions has 84 GitHub stars and 11 forks. GitHot measured 0 stars in 24 hours. Last checked 2026-11-18 11:36 UTC. Last push 2026-11-14.

How these counts are measured

Measured star growth

in 24h
0
in 7 days
Collecting history
in 15 days
Collecting history
in 1 month
Collecting history

Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.

About

A collection of tools and filters for Open WebUI.

AI

README.md

Original

Open WebUI Extensions

日本語版 / Japanese

A collection of tools and filters for Open WebUI.

🌟 Highlights

Sub Agent Tool (openwebui.com) - #1 most upvoted on openwebui.com with 20,000+ downloads! Featured in Open WebUI's official Community Newsletter, January 28th 2026 as one of "This Week's Most Useful".

Delegate tool-heavy tasks to sub-agents running in isolated contexts, keeping your main conversation clean and efficient. Fully leverages Open WebUI v0.7+ built-in tools (web search, memory, knowledge bases, etc.).

What's New

  • v0.6 — Automatic context compaction and large-result previews with selective read-back for sub-agents. Requires Open WebUI 0.9.6+ (previously 0.7.0+). Details
  • v0.5 — MCP servers configured in Open WebUI are directly available to sub-agents — no mcpo proxy needed (#6)
  • v0.4.5 — Open Terminal tools (Open WebUI v0.8.6+) are automatically forwarded to sub-agents
  • v0.4 — Skills introduced in Open WebUI v0.8 are automatically propagated to sub-agents (experimental)
  • v0.3 — Native parallel sub-agent execution via run_parallel_sub_agents

Full changelog: commits on sub_agent.py

[!TIP] If parallel execution causes issues (e.g., search API rate limits), reduce MAX_PARALLEL_AGENTS in Valves, or comment out the run_parallel_sub_agents method to disable it entirely.

Sub-agent context compaction

Tool-heavy tasks can fill the sub-agent's own context with history and results. Starting in v0.6, two mechanisms reduce the context sent to the model to help long-running tasks keep going. Both are enabled by default and can be configured independently in Valves.

  • Automatic history compaction — When estimated input tokens reach the threshold (default: 80,000), older rounds are summarized while the task and the most recent rounds are kept. By default, summaries use the same model and reuse the previously sent prefix and tool definitions, with care taken to preserve prompt caching where possible.
  • Large tool result previews — Large results (by default, 10,000 tokens or more, or over 64 KiB) are sent as a short head/tail preview instead of the full text. The original text is retained during the same task, and the sub-agent can read back the lines or ranges it needs through agent_ref_exec, using commands such as head, tail, sed -n, and grep.

Open WebUI's built-in chat-history compaction does not apply to this tool's internal sub-agent loop. This feature covers that internal history.

Valve Purpose
ENABLE_CONTEXT_COMPACTION Enable or disable automatic history compaction (default: on)
CONTEXT_COMPACTION_TOKEN_THRESHOLD Estimated input tokens at which compaction starts (default: 80,000)
COMPACTION_SUMMARY_MODEL Model used for summaries; leave empty to use the sub-agent's model (recommended)
LARGE_TOOL_RESULT_MODE ref_exec (preview + read-back, default) / truncate (omit the middle, no read-back) / raw (send as is)
LARGE_TOOL_RESULT_THRESHOLD_TOKENS Token threshold for large results (default: 10,000)
MAX_ITERATIONS Iteration cap (default: 50, previously 10; 0 means unlimited)

Parallel Tools (openwebui.com) - Featured in Open WebUI's official Community Newsletter, March 17th 2026 as one of the "Editor's Picks".

Execute multiple independent tool calls in parallel for faster execution.

Tools

Tool Description
Sub Agent Delegate tasks to autonomous sub-agents to keep context consumption low
Parallel Tools Execute multiple independent tool calls in parallel for faster results (note: often requires a strong flagship model to invoke correctly)
Multi Model Council Run a multi-model council decision with majority vote
LLM Review Multi-persona creative writing that preserves divergent voices — each persona produces a distinctive draft through independent revision and peer feedback, rather than merging into one (inspired by arXiv:2601.08003)
User Location Get user's current location via the browser's Geolocation API
Universal File Generator (Pandoc) Generate files in various formats using Pandoc. Not intended for use by others at this time

Filters

Filter Description
Current DateTime Injector Append the current datetime to the latest user message. Unlike a built-in system prompt variable such as {{CURRENT_DATETIME}}, this does not change the system prompt on every request; the aim is to avoid breaking prompt caching (a system role is available in Valves)
User Info Injector Append user info to the latest user message (same reason as above)
Full Context Mode Toggle Batch toggle full context mode per chat (the built-in feature only supports per-file toggling)

Graphiti Memory (Submodule)

A knowledge-graph-based memory extension powered by Graphiti.

Managed in a separate repository: open-webui-graphiti-memory. Referenced here via the graphiti/ submodule.

Setup

Initialize Submodules

After cloning this repository, initialize the submodules:

git submodule init
git submodule update

Or clone with submodules in one step:

git clone --recurse-submodules https://github.com/Skyzi000/open-webui-extensions.git

License

MIT License