Key concepts¶
This page defines the core terms and concepts you'll encounter throughout Aunsight Data Services. Understanding these will help you navigate the platform and its documentation more effectively.
Organizations and projects¶
Organization — The top-level container for a single client or business entity within Aunsight. All users, projects, and data belong to an organization. Aunalytics itself also has an organization for internal use.
Project — A container within an organization for a specific body of work. Projects come in two types:
- Solution project — an installation of a predefined industry data solution
- Custom project — a freeform workspace for ad hoc pipeline development
Context — The organization or project you are currently working within. Permissions are evaluated relative to the current context.
Solutions¶
Solution — A packaged, versioned, installable data product built for a specific industry use case. Solutions define a standard input model, processing logic, and deliverable output that can be deployed repeatedly across multiple clients.
Solution installation — A deployed instance of a solution within a client's project. Each installation maps the client's source data to the solution's input model.
Draft project — The development version of a solution installation where changes are made and tested before being promoted to production.
Published project — The live, production version of a solution installation.
Input model — The standardized set of tables a solution expects as input. Client source data is mapped to this model during implementation.
Processed model — The intermediate tables produced by the solution's transformation logic after the input model is populated.
Deliverable model — The final output tables produced by the solution, ready for consumption by downstream applications and users.
Module — An optional analytics add-on that can be enabled within a solution installation. Modules have configurable arguments and extend the solution's default processing.
Solution action — A unit of solution-specific logic packaged within the solution and invokable as a task. Actions can perform operations that require protected system access (such as Milvus or Bedrock), encapsulate resource-intensive workloads, or produce streaming output for AI and agentic workflows. Each action declares its argument and result structures, enabling validation throughout the call chain.
Data acquisition¶
Sync Agent — A lightweight software component installed within a client's own infrastructure. It acts as the bridge between the client's source systems and Aunsight, enabling secure data transfer without exposing internal systems directly.
Plugin — A connector that enables the Sync Agent to communicate with a specific type of data source, such as SQL Server, PostgreSQL, or an LFS bucket.
Connection — A configured link between a plugin, a Sync Agent, and a specific source system. Connections are created in the Connect tab of a solution project.
Ingestion task — A task that pulls data from a source system into the platform via a connection. Ingestion tasks can be scheduled or triggered manually.
Bucket — A storage container for unstructured data in the form of files and folders. Analogous to a file system or object store.
Local bucket — A bucket managed within an Aunsight compute environment, backed by S3-compatible storage. Used to store unstructured data acquired from client systems for downstream processing.
Remote bucket — A bucket managed by an external system, accessed via a Sync Agent plugin.
User bucket — The ability for users or pipeline actions to provide or receive files to and from a local bucket via client-side tooling.
Data storage and processing¶
Warehouse — A storage container for structured data in the form of tables and views, queryable via SQL. Warehouses come in three types:
- Local warehouse — managed within an Aunsight compute environment
- Remote warehouse — connected to an external database system
- Accelerated warehouse — optimized for high-performance, concurrent querying by downstream applications
User warehouse — A warehouse scoped to a specific user, used for personal or exploratory work within a project.
Table — A structured dataset stored within a warehouse, made up of rows and columns.
View — A virtual table defined by a query, stored within a warehouse.
Dataflow — A configured transformation that moves and reshapes data between tables. Dataflows are the primary building block of data processing pipelines. They are composed of:
- Dataflow operations — the individual steps within a dataflow that perform specific transformations such as filtering, joining, aggregating, or renaming data
- Dataflow expressions — formulas and logic used within operations to compute values, apply conditions, or derive new fields
Workgraph — The pipeline orchestration layer that ties dataflows, ingestion tasks, solution actions, and other components together into a complete, schedulable pipeline. The workgraph defines all possible work in a solution installation as a directed acyclic graph (DAG).
Tasks, jobs, and submissions¶
Task — An asynchronous unit of work executed within an Aunsight compute environment. Tasks have a well-defined lifecycle: QUEUED → RUNNING → SUCCEEDED | FAILED | CANCELLED.
Artifact — The output produced by a task. Artifacts are stored in the compute environment and retrieved directly from the Compute Gateway.
Submission — A selection from the workgraph that has been submitted for execution. A submission is the logical representation of work intended to be performed. Because a submission may not always succeed on the first attempt, it can have multiple associated runs.
Run — A single attempt to execute the work defined in a submission. A submission begins with an initial run and may have additional recovery runs if the first attempt fails.
Run recovery — The process of retrying a failed submission. A recovery run can target the full original scope, just the remaining incomplete work, or a partial subset chosen by a supporter.
Quality gate — A special workgraph node that evaluates data quality rules before allowing downstream work to proceed. If critical rules are violated, downstream processing and delivery are halted to prevent unreliable data from being delivered.
Stored run — A pre-configured subset of the workgraph that can be scheduled or triggered on demand, representing a common execution pattern for a solution installation.
Compute environment — An isolated execution environment within which tasks run and local storage is provisioned. Each environment hosts multiple client tenants and is designated for production or non-production workloads.
Users and access¶
User — A registered person who uses the platform. Users belong to an organization and can be granted access to one or more projects.
Member — A user who has been granted access to a specific organization or project. Membership determines what a user can see and do within that context.
Service user — A pseudo-user managed by an organization, used to run production workloads independently of individual users.
Role — An arbitrary collection of policies and permissions within a context. Users can be assigned roles to grant them access to specific capabilities.
Group — An arbitrary collection of roles within a context. Users placed in a group inherit all the group's permissions.
Policy — A system-defined collection of permissions that are typically needed together.
Permission — A flag that controls whether a user can perform a specific action within a context.
API key — A credential that allows programmatic access to the Aunsight API, used by the Toolbelt, SDKs, and external scripts.
Consumption¶
Consumption API — A secure API that solution-specific applications use to query warehouse data, run actions, and access installation metadata.
Consumption application — A registered application that has been granted access to specific data containers within a solution installation via the Consumption API.
Data scope — A restriction applied to a warehouse that limits which tables, views, fields, or rows are accessible to a given application or user.
IDW (Intelligent Data Warehouse) — Aunalytics' approach to delivering structured, semantically rich data models that are ready for both traditional analytics and AI/LLM applications.
MCP server — An MCP-compatible interface exposed by the Toolbelt that allows AI workflow frameworks and agents to interact with the Consumption API using standard tool-use patterns.