Data study · The open-source tooling around AI search
Open-Source AI Search Projects Drew 111,509 GitHub Stars in 2025, and Deep Research Agents Took 72.9% of Them
We tracked 17 leading open-source answer engines and deep research agents. Perplexity-style answer engines peaked in 2024, while research agents that read dozens of web pages per question took over in 2025.
Analysis of 17 open-source AI answer engine and deep research repositories, Jan 2023 to Oct 2026 · GH Archive via BigQuery, October 2026
Key findings
- 1
The 17 projects collected 214,293 recorded GitHub stars from January 2023 to October 20, 2026.
- 2
In 2025 they drew 111,509 stars. Deep research agents took 81,235 of those (72.9%), up from 28.3% of the group's stars in 2024.
- 3
Answer engines peaked in 2024 at 758.2 stars per million GitHub stars, then fell to 559.2 in 2025 and 209.7 in 2026.
- 4
Deep research agents rose from 299.2 stars per million in 2024 to 1,500.4 in 2025, five times as much.
- 5
Together the 17 projects drew 2,059.5 of every million GitHub stars in 2025, nearly double their 1,057.4 share in 2024.
- 6
Release activity moved the same way: answer engines published 141 releases in 2024 and 70 in 2025, while research agents went from 36 to 86.
Summary
Picture a plumbing company owner who types 'best water heater installer near me' into a new AI search app his nephew recommended. The app is not Google, ChatGPT or Perplexity. It is an open-source answer engine that someone set up in an afternoon. Tools like that pull web results, read the pages and write an answer with citations.
Open-source answer engines copy the Perplexity model: search the web, read the top pages, summarize. Deep research agents go further and read dozens of pages to write a full report. Anyone can run either kind, which means more places where your business can be found or missed.
We tracked 17 of the most starred projects. In 2025 they drew 111,509 recorded stars, and 72.9% went to deep research agents. Developer interest moved from quick answers to long, multi-source research.
What we measured
We used GH Archive, the public log of GitHub events, in Google BigQuery. We picked the most starred open-source answer engines and deep research agents and followed each by its permanent repo ID. We counted stars, forks, pull request events and releases by half year.
GH Archive recorded far fewer stars after mid-2025. To compare periods fairly we divided each group's stars by all stars GH Archive recorded in the same period. Project names appear only in adoption counts. We make no claims about how any project works internally.
Deep research agents overtook answer engines in 2025
Stars per million of all recorded GitHub stars, by year.
Stars per million
| Year | Answer engines | Deep research agents | Combined |
|---|---|---|---|
| 2023 | 51.8 | 80.2 | 132.1 |
| 2024 | 758.2 | 299.2 | 1,057.4 |
| 2025 | 559.2 | 1,500.4 | 2,059.5 |
| 2026 (to Oct 20) | 209.7 | 1,027.7 | 1,237.4 |
Source: aisearchoptimization.net analysis of 17 open-source AI answer engine and deep research repositories, Jan 2023 to Oct 2026 · GH Archive via BigQuery, October 2026
Answer engines had their moment in early 2024, when several Perplexity alternatives launched within weeks of each other. In early 2025 a wave of deep research agents appeared and drew 1,779.7 stars per million in the first half of the year alone.
A few projects hold most of the attention
Recorded stars by project and year.
| Project | Type | 2023 | 2024 | 2025 | 2026 (to Oct 20) | Total |
|---|---|---|---|---|---|---|
| Khoj | Answer engine | 3,819 | 12,698 | 13,391 | 780 | 30,688 |
| STORM | Deep research | 0 | 13,328 | 12,273 | 258 | 25,859 |
| Perplexica | Answer engine | 0 | 16,974 | 7,182 | 1,241 | 25,397 |
| DeerFlow | Deep research | 0 | 0 | 14,940 | 9,108 | 24,048 |
| GPT Researcher | Deep research | 5,912 | 8,970 | 7,478 | 717 | 23,077 |
| deep-research (open-source agent) | Deep research | 0 | 0 | 16,270 | 220 | 16,490 |
| Tongyi DeepResearch | Deep research | 0 | 0 | 9,612 | 352 | 9,964 |
| Search with Lepton | Answer engine | 0 | 7,787 | 293 | 12 | 8,092 |
| Morphic | Answer engine | 0 | 6,236 | 1,477 | 141 | 7,854 |
| Local Deep Researcher | Deep research | 0 | 0 | 7,035 | 175 | 7,210 |
| Open Deep Research (LangChain) | Deep research | 0 | 0 | 6,289 | 381 | 6,670 |
| Scira | Answer engine | 0 | 0 | 6,258 | 146 | 6,404 |
| MindSearch | Answer engine | 0 | 5,367 | 972 | 58 | 6,397 |
| LLM Answer Engine | Answer engine | 0 | 4,591 | 263 | 17 | 4,871 |
| node-DeepResearch | Deep research | 0 | 0 | 4,322 | 45 | 4,367 |
| Local Deep Research | Deep research | 0 | 0 | 3,016 | 581 | 3,597 |
| Farfalle | Answer engine | 0 | 2,850 | 438 | 20 | 3,308 |
Source: aisearchoptimization.net analysis of 17 open-source AI answer engine and deep research repositories, Jan 2023 to Oct 2026 · GH Archive via BigQuery, October 2026
The five largest projects hold 129,069 of the 214,293 stars (60.2%). Several answer engines that drew thousands of stars in 2024 added only a few hundred in 2025. Most deep research agents launched in the first half of 2025.
Builders followed the attention
Forks and releases show where developers are actually building.
| Year | Answer engine forks | Research agent forks | Answer engine releases | Research agent releases |
|---|---|---|---|---|
| 2023 | 179 | 703 | 33 | 11 |
| 2024 | 6,819 | 2,750 | 141 | 36 |
| 2025 | 3,785 | 8,531 | 70 | 86 |
| 2026 (to Oct 20) | 328 | 1,592 | 4 | 16 |
Source: aisearchoptimization.net analysis of 17 open-source AI answer engine and deep research repositories, Jan 2023 to Oct 2026 · GH Archive via BigQuery, October 2026
Research agents drew 8,531 forks in 2025, more than three times their 2024 count. Answer engine forks fell 44.5% over the same year. Raw 2026 numbers are low partly because GH Archive captured fewer events that year.
What this means
AI search is not one product. Besides the big assistants, anyone can run open-source answer engines and research agents, and these tools typically pull from web search results and read the top pages. A deep research agent may read 20 or 30 pages for one question, so being one of those pages matters.
Clear, factual pages that answer specific questions are more likely to be picked up by these tools than vague marketing copy. To see which of your pages these tools cite, run our free AI visibility audit; it tests real prompts and gives your LLM optimization a clear starting list.
Limits
The project list is our selection of the most starred open-source answer engines and deep research agents. Other projects exist and are not counted.
Stars reflect developer interest, not end users. Many people use hosted versions of these tools that leave no trace on GitHub.
GH Archive recorded far fewer star, fork and release events from mid-2025 onward, with gaps in October 2025 and August to October 2026. We use per-million shares for comparisons, but raw 2026 counts are understated.
One research agent accounts for a large share of the group's pull request events, which may include automated contributions, so we do not lean on pull request counts.
Embed this research
Add the charts from this study to your page. The embed links back here.
<iframe src="https://aisearchoptimization.net/embed/data-studies/open-source-ai-answer-engines-github-stars" title="Deep Research Agents Took 72.9% of AI Search Repo Stars" width="100%" height="720" style="border:0" loading="lazy"></iframe>
Cite this study
APA
Grenyo, K. (2026). Deep Research Agents Took 72.9% of AI Search Repo Stars. AI Search Optimization. https://aisearchoptimization.net/resources/data-studies/open-source-ai-answer-engines-github-stars
Plain link
Deep Research Agents Took 72.9% of AI Search Repo Stars - AI Search Optimization - https://aisearchoptimization.net/resources/data-studies/open-source-ai-answer-engines-github-stars
Frequently asked questions
What are open-source AI answer engines?
They are self-hostable tools that search the web, read the top results and write a cited answer, similar to Perplexity. The 8 we tracked drew 93,011 recorded GitHub stars from 2023 to October 2026.
What is a deep research agent?
A tool that searches and reads many web pages to write a longer report. The 9 research agents we tracked drew 81,235 recorded stars in 2025 alone, 72.9% of the group's stars that year.
Is interest in open-source AI search growing?
Yes. The 17 projects drew 1,057.4 stars per million GitHub stars in 2024 and 2,059.5 per million in 2025. Answer engines cooled while research agents grew five times over.
Why does this matter for a local business?
Each of these tools is another place customers can ask for a recommendation. They typically read top web results, so clear pages that answer real questions are more likely to be used.
Related studies
- The open-source tooling around AI searchGEO and AI SEO Repos on GitHub Hit 322 a Month in 2026
- The open-source tooling around AI searchAI Browser Agent Repos on GitHub Grew 22-Fold Since 2024
- The open-source tooling around AI searchSEO Data MCP Servers on GitHub More Than Doubled in 2026
- The open-source tooling around AI search16.1% of New SEO Repos on GitHub Now Mention AI
See where AI search leaves you out
Run the free AI visibility audit. If you want the fixes done for you, the plan is a flat $299/month.