A collection of security-focused functions and pipelines for OpenWebUI, designed to enhance security team operations.
Note
This repository includes ready-to-use pipelines and functions plus visual demos. Use the callouts to find important details quickly.
Function / Pipeline | Status | Description |
---|---|---|
TokenGuard | Proof of Concept (v0.0.2) | Experimental token management utility |
LlamaGuard Filter | Production Ready (v0.0.2) | Content filtering and safety analysis |
Anthropic Manifold Pipe | Production Ready (0.3.17) | Security analysis pipeline for Anthropic |
Anthropic Manifold Pipe Sonnet 3.7 | Beta (v0.2.x) | Enhanced capabilities, under testing |
Cloudflare WorkerAI | Beta (v0.1.6) | Early release pipeline using WorkerAI |
ThreatConnect IOC Enrichment | Production Ready (v0.7) | Indicator enrichment using ThreatConnect |
Always review code obtained from obscure repos and run at your own risk.
- Repo layout: see "Repository Structure" below
- Pipelines: manifoldPipe/
- Functions: function/
- Tools: tools/
Located in the manifoldPipe/
directory.
- File: manifoldPipe/anthropic.py
- Status: Production Ready (0.3.17)
- Purpose: General-purpose security analysis pipeline integrating with Anthropic models.
Tip
If you want to experiment with newer Anthropic features, consider the Sonnet 3.7 beta pipeline (manifoldPipe/beta-anthropic.py) in a separate workspace to avoid affecting production runs.
- File: manifoldPipe/cloudflare.py
- Status: Beta (v0.1.6)
- Purpose: Integrates Cloudflare WorkerAI capabilities into OpenWebUI pipelines.
Note
The community plugin slug may show as “cloudfare” on some community pages, but the provider is Cloudflare.
Located in the function/
directory. Copy the .py contents into OpenWebUI -> Workspaces -> Functions to add them to your instance.
- File: function/tokenGuard.py
- Status: Proof of Concept (v0.0.2)
- Purpose: Experimental token management and related utilities.
Caution
TokenGuard is experimental. Use only in test/dev environments and audit the code before any production usage.
- File: function/llama_guard.py
- Status: Production Ready (v0.0.2)
- Purpose: Content filtering and safety enforcement using LlamaGuard.
Important
LlamaGuard requires a compatible local model runtime (for example, Ollama + LlamaGuard model). Configure safety categories (S1–S13) according to your policies.
- File: function/threatconnect_ioc_enrichment.py
- Status: Production Ready (v0.7)
- Purpose: Automatically extract and enrich Indicators of Compromise (IOCs) using ThreatConnect APIs.
Note
The function/threatconnect_ioc_enrichment.py
file contains formatting helpers such as format_indicator_v2
that normalize ThreatConnect v2 responses into fields (tc_id, owner, first_seen, last_modified, confidence, rating, threat_rating, threat_confidence). It also contains authentication helper(s) like generate_tc_auth_headers
— review that function to understand authentication header construction for ThreatConnect API calls.
Tip
Open function/threatconnect_ioc_enrichment.py
to confirm which ThreatConnect API fields are mapped and to extend mappings if you need additional fields in the LLM output.
Located in the tools/
directory.
- File: tools/endgame_web_scrape.py
- Status: Beta (v0.1.2)
- Purpose: Improved web scraping with batching, retries, and cleaning options.
Note
The scraper implements robust HTTP retries with exponential backoff. See the _http_get
function in tools/endgame_web_scrape.py
for details on handling 429/503 responses, honoring Retry-After headers, and retry logic.
Tip
Use the "Clean Content" option for long articles to reduce token consumption while preserving essential context.
- Open your OpenWebUI instance.
- To add a function:
- Navigate to Workspaces -> Functions -> New Function (+)
- Copy the contents of the desired
.py
file fromfunction/
and paste into the new function editor. - Save.
- To add a manifold pipeline:
- Import or recreate the pipeline under Manifold Pipelines, copying code from
manifoldPipe/*.py
as needed.
- Import or recreate the pipeline under Manifold Pipelines, copying code from
- For tools, you can reuse scripts or adapt them into functions/pipelines.
Important
Pipelines live under “Manifold Pipelines” and must be added separately from Functions.
./
├── function/
│ ├── llama_guard.py
│ ├── threatconnect_ioc_enrichment.py
│ └── tokenGuard.py
├── manifoldPipe/
│ ├── anthropic.py
│ ├── beta-anthropic.py
│ └── cloudflare.py
├── tools/
│ └── endgame_web_scrape.py
├── asset/
│ ├── LlamaGuard_customization-hq.gif
│ ├── LlamaGuard-hq.gif
│ ├── tim_cti_analyst.png
│ ├── tokenGuard.gif
│ └── web_scrape.png
├── LICENSE
└── README.MD
-
TokenGuard
- Use: experiments and local testing
- Risk: Proof of Concept — do not rely on for production auth
-
LlamaGuard Filter
- Use: content moderation and safety enforcement
- Notes: Requires local LlamaGuard-compatible model runtime
-
ThreatConnect IOC Enrichment
- Use: IOC extraction and enrichment
- Notes: Requires ThreatConnect API credentials; inspect
function/threatconnect_ioc_enrichment.py
(includingformat_indicator_v2
andgenerate_tc_auth_headers
) to understand authorization and response formatting
-
Anthropic Manifold Pipe
- Use: production-grade analysis with Anthropic models
- Notes: Test in your environment before broad adoption
-
Cloudflare WorkerAI Pipeline
- Use: WorkerAI-based pipelines, beta — best for non-critical tasks
-
Enhanced Web Scrape Tool
- Use: Web content extraction for enrichment and context
- Notes: See
tools/endgame_web_scrape.py
for retry and rate-limit handling logic
Contributions welcome. Please:
- Open an issue for bugs or feature requests.
- Submit a pull request with focused changes, tests, and documentation updates.
I will synchronize changes to OpenWebUI community plugins manually once merged.
Created and maintained by Christian Taillon ☕
Warning
These tools augment workflows and do not replace a comprehensive security program.
- Proof of Concept (PoC) functions may contain bugs or security flaws.
- Beta functions have limited testing and may contain unknown issues.
- Even "Production Ready" functions should be validated in your environment.
Caution
Conduct your own security reviews and audits before deploying any component in sensitive environments. The author and contributors are not responsible for incidents resulting from the use of these tools.