Acceldata adds AI observability to xLake platform

Acceldata has added AI Observability to its xLake Data & AI Platform, designed to monitor and govern AI applications across on-premises systems and cloud environments.

The new offering builds on the platform's existing data quality, lineage and pipeline monitoring by adding tracing for large language model and agent-based applications. Teams can track prompts, model calls, tool use, retrieval steps and agent activity in a single view, alongside the condition of the data those systems rely on.

The launch reflects a broader shift in how companies manage AI systems in production. Rather than treating AI monitoring as separate from data oversight, suppliers are increasingly linking model behaviour to the quality, movement and governance of the underlying data.

Acceldata argues that this link matters most in hybrid technology estates, where data often remains spread across several cloud platforms and on-premises systems. In that setting, tracing an AI failure only at the application or model level may not reveal whether the root cause lies in the data product, the pipeline or the supporting infrastructure.

"The top blocker to enterprise AI is fear of ungoverned agents and data. Acceldata removes that blocker with the introduction of AI Observability to the xLake Data & AI Platform," said Rohit Choudhary, Founder and Chief Executive Officer of Acceldata. "Now you can govern and trace AI wherever your data lives, on-premises or across clouds, which lets you build agents against data you could never centralise. Governance becomes the reason you can do more with your data, not less."

One console

A central part of the product is the ability to connect AI incidents to data operations in the same console. Users can follow a failing trace back to the data source or pipeline that produced it, rather than stopping at the model layer.

Acceldata positions this approach against tools focused mainly on application infrastructure, a single cloud warehouse or a specific model environment. The platform is intended to operate across mixed estates that combine public cloud and on-premises systems.

That mixed environment is common in large organisations. Acceldata cited a GLG survey of 40 C-level leaders at enterprises with revenue above USD $5 billion, which found that 80% run hybrid architectures and 75% use four or more data platforms at the same time.

"Tracing and evaluating agents is becoming standard; what matters is what a failure connects to," said Ashwin Rajeeva, Co-Founder and Chief Technology Officer of Acceldata. "We connect it to data quality, pipeline health, lineage and compute across a hybrid estate, on-premises and cloud, on the platform teams already run for data observability."

Governance pressure

The product enters a market shaped by rising concern over AI governance in large companies. As more businesses deploy agent-based systems that can retrieve information and take actions, executives face pressure to show not just that those systems work, but that their decisions can be reconstructed and controlled.

Acceldata pointed to McKinsey research indicating that only about one-third of organisations report mature governance for agentic AI. It argues that the hardest failures to manage are those in which teams cannot reconstruct the workflow because key steps were never logged.

Its answer is to combine three functions on one platform: observation of agent behaviour, observation of the data involved, and governance in the environment where the agent runs. The aim is to reduce blind spots in cases where a team can detect a failure but cannot trace its cause, or can define a policy but cannot enforce it at runtime.

Feature set

The software includes full execution tracing structured around threads, traces and spans, allowing teams to inspect an individual AI run step by step. It also includes online evaluation of production traffic and offline evaluation for regression testing and model comparison on curated datasets.

These evaluations can use large language models as judges, as well as heuristic and custom metrics, including hallucination, answer relevance and context precision. Alerts can be triggered when quality, latency, cost or reliability passes a threshold.

Other functions include token usage and cost reporting by project, model and application version, as well as human feedback loops that let teams annotate traces and turn production runs into evaluation datasets. The software also includes sensitive data and personally identifiable information detection, configurable masking, policy checks captured in the trace, audit records of AI usage and multi-tenant isolation.

The product supports integrations with LangChain, LangGraph, LlamaIndex, CrewAI, AutoGen, ADK, OpenAI and Anthropic, alongside software development kit instrumentation for custom applications. It is being delivered as a native part of the xLake platform rather than as a separate standalone product.

Founded in 2018, Acceldata sells data management and observability software to large enterprises. Its customer roster includes Dun & Bradstreet, PubMatic, PhonePe and HCSC.