Enterprise Deep Research: Steerable Multi-Agent Deep Research for Enterprise Analytics

TL;DR

Enterprise Deep Research (EDR) presents a steerable, multi-agent framework for enterprise analytics that orchestrates adaptive planning, domain-specific retrieval, MCP-enabled tooling, visualization, and a reflection loop with optional human guidance to produce transparent, evidence-backed reports. The Master Research Agent coordinates adaptive query decomposition and context-aware planning, while specialized search agents and domain tools handle broad to enterprise-specific data sources. A persistent Research Todo Manager and a steering mechanism enable fine-grained user intervention and dynamic re-prioritization, with a three-stage result synthesis pipeline and rigorous citation management. Evaluations on open benchmarks and internal enterprise use show state-of-the-art performance and efficient resource use, complemented by a public trajectory dataset that facilitates reproducibility and further research. Overall, EDR demonstrates a scalable, auditable approach to enterprise deep research, bridging autonomous reasoning with human-in-the-loop guidance for reliable, domain-aware insights.

Abstract

As information grows exponentially, enterprises face increasing pressure to transform unstructured data into coherent, actionable insights. While autonomous agents show promise, they often struggle with domain-specific nuances, intent alignment, and enterprise integration. We present Enterprise Deep Research (EDR), a multi-agent system that integrates (1) a Master Planning Agent for adaptive query decomposition, (2) four specialized search agents (General, Academic, GitHub, LinkedIn), (3) an extensible MCP-based tool ecosystem supporting NL2SQL, file analysis, and enterprise workflows, (4) a Visualization Agent for data-driven insights, and (5) a reflection mechanism that detects knowledge gaps and updates research direction with optional human-in-the-loop steering guidance. These components enable automated report generation, real-time streaming, and seamless enterprise deployment, as validated on internal datasets. On open-ended benchmarks including DeepResearch Bench and DeepConsult, EDR outperforms state-of-the-art agentic systems without any human steering. We release the EDR framework and benchmark trajectories to advance research on multi-agent reasoning applications. Code at https://github.com/SalesforceAIResearch/enterprise-deep-research and Dataset at https://huggingface.co/datasets/Salesforce/EDR-200

Figures (6)

• Figure 1: Overview of Enterprise Deep Research framework---orchestrated system combining planning, specialized search, extensible enterprise tools, visualization, and reflection with optional human steering.
• Figure 2: Screenshot of EDR's home screen UI. This is the initial screen where users can submit their research query and add optional data files (e.g., images, sheets, databases). The research offers 3 run modes: quick (for a high-level investigation), standard (the standard deep research mode), and deep (for a max-effort report) and allows the user to choose their underlying base LLM.
• Figure 3: Screenshot of EDR's research execution screen. Left pane shows the progress with a summary of the current step, while the progress pane shows the todo status. Each task is annotated with its status and provenance with timestamps for full traceability. The received steering message is queued until the reflection phase.
• Figure 4: Prompt for generating the initial research task plan with 3–5 prioritized tasks guiding the first research loop.
• Figure 5: Prompt defining the query decomposition and execution process for targeted, time-sensitive, and context-aware research queries.