derisk-ai/OpenDerisk

OpenDeRisk – AI‑native Risk Intelligence System

What it is – OpenDeRisk is an open‑source platform that uses multiple AI agents to automate root‑cause analysis (RCA) for software reliability (SRE). It ingests large‑scale incident data (the Microsoft OpenRCA dataset) and, through coordinated agents, extracts the underlying cause, visualises the evidence chain, and produces reports.

Key capabilities

1. DeepResearch RCA – parses logs, traces and code to pinpoint the true source of a failure.
2. Evidence‑chain visualisation – interactive UI shows the full diagnostic flow so engineers can verify the reasoning.
3. Multi‑agent collaboration – specialised agents (SRE‑Agent, Code‑Agent, Report‑Agent, Vis‑Agent, Data‑Agent) work together; the Code‑Agent can even generate analysis scripts on the fly.
4. Open architecture – all components are modular and reusable, with optional extensions (e.g., DingTalk notifications, ChromaDB storage).

How it works

• Data layer pulls and decompresses the ~26 GB OpenRCA dataset, making it available for analysis.
• Logic layer runs the agent swarm. Each agent has a defined role and communicates via a shared protocol (MCP – Model Context Protocol).
• Visualization layer uses the Vis protocol to render the processing pipeline and evidence chain in the web UI.

Getting started

1. Quick install – run the one‑line curl script which installs the binary and a default config file.
2. Configuration – edit ~/.openderisk/configs/derisk-proxy-aliyun.toml to add your LLM API keys.
3. Run – openderisk‑server (or uv run derisk quickstart) starts a local server; the UI is reachable at http://localhost:7777.
4. From source – install the lightweight Python package manager uv, clone the repo, and sync dependencies with the provided extras (base, proxy_openai, rag, storage_chromadb, etc.).

Usage modes

• AI‑SRE (OpenRCA) – analyse telecom‑grade incident datasets.
• Flame‑graph assistant – upload Java/Python flame graphs for automated performance diagnosis.
• DataExpert – drop in metrics, logs, traces or Excel files and converse with the system to explore the data.

Extensibility – developers can add new agents under derisk‑ext.agent.agents or create custom tools/skills (MCP, DeRisk‑Skills). The project reuses ideas from DB‑GPT, GPT‑Vis, MetaGPT and Microsoft’s OpenRCA.

Community – Discord, DingTalk, and a public GitHub repo with issue tracking, releases, and a citation for academic use.

---

TL;DR – OpenDeRisk is a modular, multi‑agent AI platform for automated root‑cause analysis and risk intelligence, packaged with a web UI, visual evidence chains, and support for custom extensions.