aryan-anakin/research-agent — explained in plain English
Analysis updated 2026-05-18
Research a company's ideal customer profile or any open-ended topic with cited, verified sources.
Run a multi-worker research pipeline that searches, scrapes, and verifies claims before writing a report.
Resume or inspect a past research run by reading its saved plan, findings, and sources files.
| aryan-anakin/research-agent | 0xkinno/neuralvault | 0xlocker/d17-contracts | |
|---|---|---|---|
| Stars | 1 | 1 | 1 |
| Language | — | TypeScript | Solidity |
| Setup difficulty | moderate | hard | hard |
| Complexity | 4/5 | 4/5 | 5/5 |
| Audience | vibe coder | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires Claude Code, an Anakin account with OAuth sign-in, and spends Anakin credits on every run with no dry-run mode.
Deep Research Agent is a system for automatically researching any topic, built entirely on top of Claude Code and a remote service called Anakin. Claude Code acts as the orchestrator, using its skills and subagent features plus files on disk to keep track of state, while Anakin provides the actual research capabilities such as web search and web page scraping through a hosted connection called an MCP server. You give it a research question, and it breaks the question down into around five to nine independent subtasks, then spins up a separate worker for each subtask. Each worker searches the web for candidate sources, called seeds, then scrapes the most promising pages to check whether a claim is actually backed up by the real page content. Only claims confirmed against scraped text can be cited in the final report, which is a two step process meant to reduce made up or unverifiable citations. A separate verification step double checks claims adversarially before the final report is written. Every run stores its plan, findings, sources, and final report as files inside a runs folder, so a run can be inspected, edited, or resumed later rather than existing only in the AI's temporary memory. Before it spends any resources, the tool pauses and asks for your explicit approval, since every run scrapes live web pages and spends Anakin credits, and there is no practice or dry run mode. To use it, you open the project in Claude Code, approve the Anakin connection, sign in through a browser based login, and then run a single command with your research question. It first asks where to save the results, creates a free plan, and only proceeds to the costly search and scrape steps after you confirm. Each run is capped at roughly 200 credits. The project describes itself as an early version with a single research pass working today, and lists planned future phases including a reflection step that runs a second research wave to fill gaps, and a scheduled digest of what changed about a target topic over time.
Deep Research Agent uses Claude Code plus the hosted Anakin service to research any topic, verifying claims against scraped pages before citing them.
No license information is provided in the README.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly vibe coder.
This repo across BitVibe Labs
Verify against the repo before relying on details.