What is SPARK?
SPARK is the ecosystem for agentic AI in investment operations: every use case a plugin, built in five layers, and an app in three tabs.
SPARK is the ecosystem for agentic AI in investment operations. Agents work inside your workspace, grounded in your data, following NGT Skills and calling NGT Tools for every number, with people deciding what matters. This page explains the idea behind SPARK and the three tabs you will use every day.
Every use case is a plugin
A plugin is one use case, packaged: NAV oversight, fee checks, look-through, referential data and many more, configured on your needs. Each plugin keeps the full power of the model, inside five layers that decide where it operates:
| Layer | What it does | Where you find it |
|---|---|---|
| Agent Studio | Identity, connectors, vaults, permissions and memory. Where the plugin is built. | Studio on the Ecosystem tab, and Agent Studio in Settings |
| Orchestration | A deterministic workflow or an agent runs the steps. Routine steps need no model call. | Workflows, and the agents you name in threads |
| NGT Skills | Own the procedure and call NGT Tools. Given, not invented. | Agent Studio's Skills, and the NGT Skills pages |
| NGT Tools | Deterministic calculations and connectors. Numbers fetched, never inferred. | Agent Studio's Tools, and the NGT Tools pages |
| Guardrails | A person or an agent checks the exit. Every run replayable. | Agent Studio's Guardrails, and the approvals in workflows |
NGT Tools run on three engines: NGT Data Management (any source in, harmonised data out), NGT Analytics (every number calculated, never inferred) and the NGT Exception Framework (check everything, open only what is worth your time). See NGT Tools.
Human on the loop. Agents own progress; people own decisions. Every action happens inside your workspace's permissions and stays visible, so you can trust an agent with real work instead of just suggestions.
Agents and the context around them
An agent is only as good as the context it can see, so SPARK pairs its agents with your operations:
- Agents do the work. Every workspace comes with Spark, the built-in agent that can use all of SPARK's tools, beside the built-in Solutions Engineer and Coding Assistant. You build agents of your own, with their own instructions, model, tools and skills, in Agent Studio.
- The context is what they read and act on: your data objects, canvases, tasks, files, workflows and Spaces. These are also the objects you work with directly, every day.
Your organisation's settings show where SPARK is hosted and where its data is kept (Google Cloud in the EU for SPARK's cloud), and the model calls NGT runs for you go to EU endpoints; see Hosting & data regions.
The app in three tabs
The sidebar has three tabs:
- Threads: where you talk with people and agents. A workspace opens on its New Thread page; your channels, direct messages and threads sit in the Threads panel. An agent answers when you name it.
- Ecosystem: what the workspace holds. Agent Studio, Tasks, Canvas, Workflows, Data and Files, your Spaces and your agents.
- Settings: configuration, from Agent Studio and the extensions to the organisation, the workspace and monitoring.
Core objects at a glance
| Object | What it is | Learn more |
|---|---|---|
| Workspace | Your team's tenant: data, members and settings | Organisation |
| Thread | A conversation with people and agents, in a channel or a direct message | Start a thread |
| Agent | Does the work, starting with the built-in Spark | Create an agent |
| Space | A container that packages a use case | Spaces |
| Data object | Structured data, like a table | Data |
| Canvas | A board of widgets on your data | Canvas |
| Task | Tracked work linked to your data | Tasks |
| File | Documents and inputs, including SFTP and the Mailbox | Files |
| Workflow | Triggers and actions that automate work | Workflows |
Where to go next
- New here? Follow the Quick start to get to your first result in about 10 minutes.
- Find your way around: Navigating the workspace.
- Want the vocabulary first? Skim Key concepts.