ROI Updates | 30 March 2026

Introducing Kubro(TM) Agents

A Toolkit For Bespoke Agentic Workflows In Market Research

When it comes to professional-grade market research, full control and management of the parameters used in 'Deep Research'-type AI applications will be (aside from proprietary data) a source of differentiation, in our view.

With that in mind, we are introducing Kubro(TM) Agents - a module within the Kubro(TM) platform for building and deploying bespoke agentic research workflows, based on several use case-specific pre-defined workflow templates.

The module enables data companies, investment research teams, and market intelligence operations to construct fully configurable, multi-step 'deep research' workflows that integrate open web sources and third-party tools with the Kubro(TM) system, with a built-in quality evaluation and full observability, all within a single platform.

Kubro(TM) Agents' initial pre-defined and tested workflows include Market Research and People Research templates. 

Granular Control Over the Deep Research Parameters

Deep research APIs from major AI labs are powerful, but they don't provide full control over the research process. In professional market research, the methodology matters. That's why we built Kubro(TM) Agents around full configurability.

Kubro(TM) Agents make every step of the research workflow individually configurable. With over 50 parameters available across the workflow, research teams define not just what to research, but precisely how — which sources to include or exclude, what classification logic to apply, how to structure the LLM prompts, and what quality thresholds to enforce before results are delivered. The workflows can be triggered on demand, run on a recurring schedule, or saved as reusable templates.

Configuring third-party Search APIs (Tavily, SERP)

Configuring integration with Kubro(TM)

Context engineering - configuring additional classifiers

In the evaluation, you can combine human/deterministic rules with LLM evaluation, using regular expressions.

Third-Party Integrations

You can use different LLMs for different tasks in a workflow. Currently, we can support any models via Bedrock and Azure OpenAI, Claude, Grok, and Gemini - simply configure your API key and controls, and we can add other models on request.

Model selection in deep research-type workflows can make a big difference in terms of total token cost when deployed at scale. Some parts of the workflow can be handled with lower-cost, simpler models, while some may call for the best ones.

For Web Search APIs, aside from what can already feed into Kubro(TM), you can use Tavily and SERP.

More to come soon.

Specialist Platform, Not a General-Purpose Workflow Tool

Unlike general-purpose workflow automation tools, Kubro(TM) Agents is purpose-built for market research and operates within the broader Kubro(TM) Information Engine.

Research workflows draw on Kubro's existing capabilities in automated data collection from unstructured data sources, text classification with domain-specific ontologies, relevance scoring, and structured data extraction — combining years of market research methodology with agentic AI orchestration.

'People Research' workflow, for example, focuses on a very specific set of tasks related to research on company executives:

We have been deploying agentic research workflows internally at the sister company Real Estate Foresight ("REF"), an independent research firm focused on China property markets since 2012.

Deployment

At this stage, Kubro(TM) Agents is available for Kubro(TM) clients for testing and for bespoke deployments by the ROI Team.