AI data agent
An AI data agent is software that answers analytical questions by planning and executing its own steps — querying approved databases, running code, and citing what it used — rather than only generating text.
The distinction that matters is between generating an answer and producing one. A chatbot predicts a plausible response from what it has read. An agent decides which sources to consult, issues real queries against them, inspects the results, and revises its approach when the data does not support its first attempt. The output is grounded in an execution trace rather than in recall.
That difference becomes decisive in regulated settings. A number produced by an agent can be traced back to the query that generated it, the dataset it ran against, and the permissions that allowed the access. A number produced by a language model alone cannot be traced anywhere, because no retrieval or computation took place.
Agents also operate under constraints that a chat interface does not express. Which databases are in scope, which tools may be invoked, what happens when a request exceeds the operator's authority — these are properties of the agent's runtime, not of the prompt.
What distinguishes it
- Plans and executes multi-step analysis instead of returning a single generated response
- Runs against approved data sources under explicit, enforced permissions
- Produces an execution record that can be inspected after the fact
- Operates inside a defined boundary rather than calling arbitrary external services
In DataQube
In DataQube, an agent runs inside the customer's own infrastructure, and its work accumulates in a thread — an append-only record of the conversation, the queries issued, the code executed, and the permission decisions made along the way. Nothing is reconstructed after the fact, because the record is written as the work happens.
Common questions
How is an AI data agent different from a chatbot?
A chatbot generates a response from patterns in its training data. An AI data agent plans a sequence of steps, executes them against real data sources, and reports what it found. The agent's answer is the product of retrieval and computation; the chatbot's is the product of prediction.
Does an AI data agent need access to production data?
It needs access to whichever sources the operator approves, and no more. Scope is defined by the deployment rather than by the agent, and read-only connections with enforced row limits are a common configuration in regulated environments.
Can an AI data agent run without internet access?
Yes, provided the model and runtime are hosted inside the same network. This is the air-gapped deployment pattern, and it requires that licensing and model inference both work offline.
Related terms
Data provenance
Data provenance is the recorded lineage of a result: the sources, queries, transformations, and decisions that produced it, retained in enough detail that someone else can verify the figure independently.
Air-gapped AI
Air-gapped AI is an AI system that operates entirely inside a network with no external connectivity, meaning model inference, data access, and licence validation all function without any call leaving the perimeter.
Deterministic workflow
A deterministic workflow is an ordered, parameterised sequence of steps that produces the same result when run again with the same inputs, making it suitable for recurring reporting and review processes.