Open Source · AI-Native OpenTelemetry APM
Ask one question in plain English — AI agents query metrics, traces, and topology together, then tell you what broke.
Live Demo | Documentation | 中文 | Community
Username:Admin, Password:Databuff@123
Ask one question in plain English — AI agents query metrics, traces, and topology together, then tell you what broke.
Live Demo | Documentation | 中文 | Community
Username:Admin, Password:Databuff@123
AI multi-agent troubleshooting · Service health · Call graph topology — English UI
AIOps Roadmap: Visible → Legion → Inspect → Diagnose → Repair → Predict → Answer
① Visible · Natural language query
Ask "which service was slowest in the last hour" in plain English — AI queries 20 services itself, returns the top 3 with average latency. No query language to learn.
② Legion · Multi-agent parallel dispatch
Don't pick a specific expert — hand the complex task to the AI Brain. It dispatches to the query + inspection experts in parallel, then synthesizes a forwardable incident report.
③ Inspect · Inspection + HTML report
One sentence triggers a single-service inspection. 81 seconds later, a fully formatted HTML report appears: entry health 98, downstream MySQL 60, Redis 100, active alerts 0 — the entry looks perfectly healthy, yet the error log section surfaces 60 InsufficientStockException errors in 30 minutes, hidden behind HTTP 200. Preview and forward directly.
④ Diagnose · Root cause analysis with evidence chain
Ask "is the bottleneck in the app, the database, or downstream?" — AI pulls topology, ranks outbound call metrics, attributes by share. Bottleneck is downstream service-b at 73.2%; other components are cleared. The conclusion is ready to paste into an incident report.
⑤ Repair · Ops expert SSHes in and acts
Beyond reading dashboards, it can log into the machine and fix things. Container stuck in a restart loop? The ops expert SSHes in, runs docker logs / inspect / free -m, identifies OOM 137 — the container memory limit was only 10MB while the JVM needed 64MB. It removes the old container, recreates it with 512MB, and docker ps is stable again. The watershed from "can see" to "can repair".
⑥ Predict · Capacity health analysis
From "post-incident firefighting" to "pre-incident prediction". AI clarifies the real dependency graph, then judges the high-latency Redis — 98 QPS is far below single-node 10k capacity, so the bottleneck is not capacity but big keys / slow commands. Clear advice: don't blindly scale up.
⑦ Answer · Open-source product with a built-in concierge
"How do I integrate the OTel SDK? Where do I configure alert thresholds?" The Q&A expert actually reads the product docs and returns OTLP ports (gRPC 4317 / HTTP 4318), the Spring Boot Java Agent one-liner for zero-code instrumentation, and the alert config menu path — its own docs, more reliable than a search engine.
DataBuff supports both OpenTelemetry and SkyWalking native ingestion — existing agents switch over without rework.
| Protocol | Port / Endpoint | Supported signals |
|---|---|---|
| OTLP (OpenTelemetry native) | gRPC 4317 · HTTP 4318 |
Traces + Metrics + Logs |
| SkyWalking native gRPC | gRPC 11800 |
Trace Segment + JVM metrics + Logs (reuse existing SW Agent, just change exporter address) |
AI reads the data for you; the UI confirms what the AI said. Two legs walking. Global topology, service list, service detail, service flow — every viewpoint is there, drill down from the topology to a single trace.
Service list · Traffic-light status for anomalies |
Global topology · Auto-generated call graph |
Service detail · Metric trends and instances |
Service flow · Upstream and downstream dependencies |
Open http://YOUR_HOST:27403 · Default login admin / Databuff@123 · Add your API key in model settings to enable AI
Contributing Guide · Open Issues · Discussions · Good First Issues
See CONTRIBUTING.md to learn how to submit PRs, report bugs, or request features.
Join our WeChat community for real-time help 👇
