Be Certain is a high-performance Python analytics engine designed to transform raw telemetry into actionable intelligence. By correlating metrics from Mimir, traces from Tempo, and logs from Loki, it provides deep-tier anomaly detection, predictive forecasting, and automated Root Cause Analysis (RCA).
Built with a modular, "drop-in" architecture, Be Certain allows SRE teams to utilize pre-built analysis modules or extend the engine with custom logic for specific domain needs.
Note: Be Certain is currently in active development. We are refining our ML models and welcome PRs to help stabilize core heuristics.
- 🧠 Multi-Dimensional Anomaly Detection: Detect silent failures using advanced time-series pattern recognition.
- 📈 Intelligent Forecasting: Move from reactive to proactive with predictive trajectory models and baseline bands.
- 🔗 Causal Correlation: Understand the why by linking disparate events across logs and traces.
- 📊 SLO-Centric Monitoring: Automate error-budget burn rate calculations and availability targets.
- 🔌 Plug-and-Play Connectors: Native support for the LGTM stack (Loki, Grafana, Tempo, Mimir) and VictoriaMetrics.
- 🧪 Developer First: Modular internal packages and a comprehensive
pytestsuite for reliable extensibility.
The heart of Be Certain is the POST /api/v1/analyze endpoint. It orchestrates a staged pipeline that moves from raw data ingestion to high-level hypothesis ranking.
| Stage | Responsibility | Logic |
|---|---|---|
| Orchestration | analyzer.py |
The "Conductor." Manages the workflow from fetch to final report. |
| Ingestion | fetcher.py |
High-concurrency data retrieval with smart fallback mechanisms. |
| Detection | anomaly/* |
Identifies structural shifts and classifies severity in real-time. |
| Context | baseline/* |
Computes dynamic Z-score bands to interpret normal vs. abnormal behavior. |
| Shifts | changepoint/* |
Uses CUSUM logic to detect sudden oscillations or permanent shifts. |
| Signals | logs/* & traces/* |
Extracts log patterns and maps latency degradation across service spans. |
| Logic | correlation/* |
Temporally aligns anomalies to find "clumps" of evidence. |
| Causality | causal/* |
Applies Granger-causality and Bayesian posteriors to find the source. |
| Hypothesis | rca/* |
Ranks likely root causes based on evidence weights and topology. |
| Topology | topology/* |
Maps the "Blast Radius" and upstream dependencies of a failure. |
Be Certain is structured to be "import-friendly." Whether you are running the full API or just using the engine as a library, the structure is clean and predictable.
.
├── api/ # FastAPI routes and Pydantic schemas
├── connectors/ # Client wrappers for Loki, Mimir, Tempo
├── datasources/ # Abstraction layer for multi-source data fetching
├── engine/ # The "Brain" - individual analysis modules
│ ├── anomaly/ # Detection heuristics
│ ├── causal/ # Bayesian & Granger logic
│ ├── ml/ # Clustering and scoring weights
│ └── rca/ # Root cause ranking
├── store/ # Persistence layer for results and baselines
└── tests/ # Exhaustive component & integration tests
Clone the repository into your workspace:
git clone https://github.com/observantio/becertain.git BeCertain
cd BeCertainThe easiest way to get started is using the provided Dockerfile:
docker build -t becertain:latest .
docker run --rm -it -p 8000:8000 --name becertain becertain:latestFor local testing or debugging individual modules:
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
python run.pyWe love contributors! Whether it's a new causal algorithm or a bug fix in the OTel fetcher, your help is appreciated.
Developer Checklist:
- Create a feature branch.
- Ensure all tests pass:
pytest -q. - Ensure your
pre-commithooks are active.
Licensed under the Apache License 2.0. You are free to use, modify, and distribute this software, provided that all original attribution notices are preserved.
Disclaimer: This software is provided "as is" without warranty. The maintainers are not affiliated with third-party service providers mentioned in the connectors.