Open Deep-Research Agent, Evaluation

Overview

A clone of the open-source open-deep-research project, a TypeScript agent that performs iterative, recursive web research (search, scrape, reason, expand) and synthesizes a report. Retained for evaluation; upstream history only.

Why It Exists

To evaluate a compact, self-hostable “deep research” agent: how it decomposes a question, fans out searches via Firecrawl, recursively follows leads to a configurable breadth/depth, and produces a cited synthesis.

What We Built

No custom development; an upstream snapshot studied as reference. Observed: a Node/TypeScript pipeline using the Vercel AI SDK with OpenAI (and local-LLM alternatives), Firecrawl for search and scraping, recursive query expansion, a text-splitter with test coverage, and Docker packaging.

Technologies & Approach

TypeScript on Node with the Vercel AI SDK and Firecrawl. The pattern evaluated is recursive research with breadth/depth controls and follow-up question generation feeding subsequent search rounds.

Outcome / Impact

Gave the studio a concrete reference for autonomous research-agent design (iterative search + synthesis), relevant to internal research-assistant work. Documented as evaluation/R&D.

Capabilities Demonstrated

• Evaluating autonomous deep-research agent architectures
• Understanding iterative search/scrape/synthesize pipelines with Firecrawl