WhyLabs was a privacy-preserving AI observability platform that monitored model health, detected data drift, and ensured AI system quality by creating statistical profiles of data rather than storing raw inputs. Apple acquired WhyLabs in January 2025 and discontinued the commercial platform, but the full codebase — including whylogs (2.7k+ GitHub stars) and LangKit (960+ stars) — has been released as open source. It is listed in the AI security category.
Founded by Alessya Visnjic (who joined Apple as an engineering leader after the acquisition), WhyLabs was backed by Bezos Expeditions, AI Fund, Defy Partners, and Madrona Venture Group, raising $14M across Seed and Series A rounds. The company established itself as a pioneer in the AI observability category, with a distinctive focus on privacy-preserving monitoring — creating statistical profiles of data rather than storing raw inputs.
What is WhyLabs?
WhyLabs sat at the intersection of AI monitoring and data privacy. The core idea: organizations need to monitor their AI systems continuously but often cannot log raw data due to privacy regulations, compliance requirements, or the sheer volume of production data.
The platform solved this through statistical profiling. whylogs, the underlying open-source library, creates lightweight statistical summaries of datasets — capturing distributions, missing values, type information, and drift metrics — without retaining the original data. Teams could detect problems (data drift, quality degradation, model performance drops) while staying privacy-compliant.
With the Apple acquisition, the commercial platform is no longer available, but the technology lives on through three open-source projects.
Key Features
| Feature | Details |
|---|---|
| Core Approach | Privacy-preserving monitoring through statistical profiling |
| Data Logging | whylogs — lightweight profiles without raw data storage |
| LLM Monitoring | LangKit — prompt/response quality, toxicity, relevance, safety |
| Drift Detection | Statistical comparison of data distributions over time |
| Data Quality | Missing values, type validation, distribution anomalies |
| Model Health | Performance tracking, accuracy monitoring, degradation alerts |
| Privacy | No raw data storage — only statistical summaries |
| Status | Acquired by Apple (January 2025); commercial platform discontinued |
| Open Source | Full platform, whylogs, and LangKit available on GitHub |
| License | Apache 2.0 (whylogs and LangKit) |
whylogs: Privacy-Preserving Data Logging
whylogs is the library that made WhyLabs distinctive. It creates mergeable, lightweight statistical profiles of any dataset — whether tabular data feeding a classification model or text data flowing through an LLM.
A whylogs profile captures statistical summaries (mean, median, standard deviation, quantiles), distribution information (histograms, frequent items), data quality metrics (null counts, type mismatches), and cardinality estimates. These profiles can be compared over time to detect drift — when production data starts diverging from training data or from previous time periods. Because profiles are statistical summaries rather than raw data, they can be safely stored, shared, and analyzed without privacy concerns.
LangKit: LLM Monitoring
LangKit extends whylogs specifically for LLM monitoring. It extracts signals from prompts and responses that indicate quality, safety, and relevance:
- Text quality — Readability scores, fluency metrics, coherence evaluation
- Relevance — How well responses align with the prompt context
- Sentiment analysis — Emotional tone and polarity detection
- Toxicity detection — Harmful, offensive, or inappropriate content flagging
- Safety assessment — Prompt injection indicators, sensitive content detection
LangKit is designed to work alongside whylogs, meaning all LLM monitoring data gets the same privacy-preserving treatment — statistical profiles rather than stored prompt-response pairs.
The Apple Acquisition
Apple acquired WhyLabs in January 2025 in a quiet deal disclosed through European Commission filings under the Digital Markets Act. WhyLabs co-founder and CEO Alessya Visnjic joined Apple as an engineering leader. The acquisition fit Apple’s expanding AI efforts and its need for internal AI observability.
Following the acquisition, WhyLabs discontinued its commercial platform and open-sourced the complete codebase, ensuring the technology remains available to the community.
Getting Started
pip install whylogs. It works with pandas DataFrames, Spark, and raw Python data structures. Apache 2.0 licensed.import whylogs as why; result = why.log(df). Each profile captures distributions, quality metrics, and drift baselines without storing raw data.pip install langkit. Configure it to extract quality, safety, and relevance signals from LLM prompts and responses.When to Use WhyLabs Tools
Although the commercial platform is gone, the open-source components remain valuable for specific use cases. whylogs is particularly useful for teams that need to monitor data quality and drift in privacy-sensitive environments — healthcare, finance, and any domain where logging raw production data creates compliance risks. LangKit is a lightweight, privacy-preserving alternative to commercial LLM monitoring platforms.
The trade-off is that self-hosting requires more operational effort than using a managed platform, and the open-source projects will not receive the same level of commercial development and support they had before the acquisition.
How WhyLabs Compares
WhyLabs pioneered privacy-preserving AI observability, and its open-source tools remain relevant in the AI security landscape. For actively maintained commercial AI observability, see Arize AI (OpenTelemetry-based tracing, Phoenix open-source, AX enterprise platform) or Arthur AI (model monitoring, bias detection, LLM firewall).
For LLM-specific security rather than observability, consider Lakera Guard or Prompt Security. For runtime guardrails, see NeMo Guardrails or LLM Guard.
For a broader overview of AI security tools, see the AI security tools category page.
Note: Acquired by Apple in January 2025 and subsequently closed. The full platform, whylogs, and LangKit have been open-sourced at github.com/whylabs.
