https://github.com/nblx-org/authring
Authring is an AI-driven guardrails system that evaluates code change requests using Astra DB, Langflow, Unstructured.io, and Twilio.
https://github.com/nblx-org/authring
agentic-workflow astradb automation devops guardrails langflow machine-learning twilio unstructured
Last synced: 2 months ago
JSON representation
Authring is an AI-driven guardrails system that evaluates code change requests using Astra DB, Langflow, Unstructured.io, and Twilio.
- Host: GitHub
- URL: https://github.com/nblx-org/authring
- Owner: Nblx-org
- License: mit
- Created: 2025-03-01T02:12:11.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2025-03-02T07:02:17.000Z (over 1 year ago)
- Last Synced: 2025-09-23T13:20:47.614Z (9 months ago)
- Topics: agentic-workflow, astradb, automation, devops, guardrails, langflow, machine-learning, twilio, unstructured
- Language: CSS
- Homepage: https://authring.org
- Size: 104 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Authring
[](https://pypi.org/project/authring/)
Authring is an MCP-driven guardrails system that evaluates code modification proposals based on historical impact, contributor expertise, and production risk assessment. The system utilizes:
- **Astra DB** for historical data storage
- **Unstructured.io** for document parsing
- **Langflow** for AI-driven decision-making
- **Twilio** for real-time human validation
- **GitHub CLI (gh)** for managing patch review links
## Workflow
1. **MCP Server** receives a proposed patch in the format:
```json
{
"repo": "flight-control",
"purpose": "Fix typo in logging message",
"patch": "diff --git a/utils/logger.py b/utils/logger.py\n- log.info('Engne diagnostics running')\n+ log.info('Engine diagnostics running')"
}
```
2. **Contextualization**: The system extracts metadata from `mock/code/` to determine the impact, authorship, and production status of the affected code.
3. **Risk Assessment**:
- If **Safe**, the patch is accepted.
- If **Unsafe**, the patch is rejected.
- If **Maybe**, a **Twilio notification** is sent to the responsible contributor.
4. **Twilio Review (Maybe Cases)**:
- The contributor receives a text:
```
Authring wants to 'Improve flight path prediction' by changing 'navigation.py'.
Review it here: and respond with your recommendation.
```
- The system waits for a response (15s timeout) before making a final decision.
## Demo Scenarios
- ✅ **Valid Purpose, Valid Patch**: Fix typo in `logger.py`.
- ❌ **Invalid Purpose**: Replace autopilot with `YOLO()` function.
- ⚠️ **Valid Purpose, Invalid Patch**: Change flight path prediction logic (triggers Twilio).
## Setup
```sh
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
python mcp_server.py
```
## Running the Demo
```sh
python train.py # Populate mock Astra DB with training data
python mcp_server.py # Start the evaluation pipeline
```