State of Open-Source AI on GitHub 2026

1. 1. Repository-category counts and GitHub stars are different maps

   Within this snapshot's category assignment, LLMs is the largest category by repository count (68, 20.9%) and holds 28.4% of cohort stars. Repositories assigned to MCP Servers and AI / ML hold more star share (12.7% and 15.5%) than repository share (7.7% and 7.1%). LangChain shows the reverse (12.3% of repositories, 5.1% of stars). These are DataAIHub operational categories for this snapshot, not a product taxonomy.

2. 2. Accumulated stars and recorded 30-day commits are sharply different distributions

   Within this eligible cohort, the 10 repositories with the most stars hold 14.0% of cohort star mass, while the 10 with the most recorded 30-day commits hold 65.7% of cohort recorded 30-day commit volume. The two top-10 sets overlap in only 3 repositories, so the comparison describes two different concentration patterns rather than the same ten repositories dominating both measures.

3. 3. GitHub push recency and commit volume measure different things

   227 of 326 repositories (69.6%) have a GitHub push within 30 days of the snapshot, while 135 (41.4%) recorded zero commits in that window. Neither figure is a quality judgment; the two fields measure different events.

4. 4. High star counts can sit beside little recorded commit volume

   obra/superpowers is #2 by stars (288K) in this cohort, with 0 recorded 30-day commits. Star popularity and recorded 30-day commit volume are separate signals.

5. 5. AI Agents are a large slice of this cohort, not the whole ecosystem

   63 of 326 repositories (19.3%) are assigned to the AI Agents category and hold 18.6% of cohort stars — roughly proportional to repository count.

6. 6. Python leads overall; TypeScript is concentrated in some categories

   Python is the primary GitHub language for 176 of 326 repositories (54.0%). Repositories assigned to MCP Servers have the strongest TypeScript share among the eight categories: 10 of 25 (40.0%).

All figures below are repository-level GitHub metrics for the 326 eligible repositories in this snapshot. Star and fork totals are sums across repositories. They are not counts of unique people.

135 repositories (41.4%) recorded zero commits in the 30-day window. Observed contributor counts sum to 72,231. Contributor counts are capped at 1,000; 23 repositories sit at that cap. This is not a complete contributor census.

Repository creation dates in this cohort span 2010–2026. 75 repositories were created in 2023 (23.0%); 29 were created in 2026 (8.9%). That is an age distribution of GitHub creation timestamps, not an emerging-projects ranking and not a measure of when a project became an AI codebase.

DataAIHub assigns each tracked repository one operational category for this snapshot based on repository discovery and GitHub topic signals. The eight categories are operational buckets for this snapshot, not mutually exclusive product domains. A serving engine can sit in LLMs; an automation platform with MCP topics can sit in MCP Servers.

Within this snapshot's category assignment, repository share and star share are not the same map. LLMs: 20.9% of repositories and 28.4% of stars. MCP Servers: 7.7% → 12.7%. AI / ML: 7.1% → 15.5%. LangChain: 12.3% → 5.1%. AI Agents: 19.3% → 18.6%, roughly proportional to count. Star share for a category can be dominated by a few high-star repositories and by topic-based assignment. It is not a measure of a product market.

Category assignment follows the snapshot's operational classification rules rather than a manual product taxonomy, so individual repositories may appear in categories that differ from how readers commonly describe them. For example, huggingface/transformers is represented within LLMs in this snapshot, while langchain-ai/langchain is represented within AI Agents.

The question in this section is how different accumulated GitHub popularity is from recorded 30-day commit volume in this cohort. Stars and forks measure accumulated attention. Recorded 30-day commits are derived from GitHub's four-week commit-activity data.

Within this eligible cohort, accumulated GitHub star mass is moderately concentrated, while recorded short-window commit volume is much more concentrated. The 10 repositories with the most stars hold 14.0% of cohort star mass. The 10 with the most recorded 30-day commits hold 65.7% of cohort recorded 30-day commit volume. The two top-10 sets overlap in only 3 repositories, so the comparison describes two different concentration patterns rather than the same ten repositories dominating both measures. The single highest recorded 30-day commit count is 17.9% of the cohort total (openclaw/openclaw).

GitHub push recency and recorded commit volume measure different things. 227 of 326 repositories (69.6%) have a last GitHub push within 30 days of the snapshot, while 135 (41.4%) recorded zero commits in that window. Prefer recorded 30-day commit counts when the question is short-window commit volume. Neither figure is a quality judgment.

Repositories can lead on stars while showing little recorded 30-day commit volume. The table is sorted by stars among repositories with zero recorded 30-day commits. It is not a ranking of inactivity.

A recorded zero should not be read as proof of inactivity. The repository's last-push date and the four-week commit-activity measure capture different events. A recent push therefore does not necessarily imply a non-zero recorded commit count. The underlying activity data can also be incomplete or unavailable. A recorded zero should therefore be read as a measurement result for this snapshot, not independent proof that a repository had no development activity.

The opposite mismatch also appears: high recorded 30-day commit volume beside a much lower star rank. The rows below are among the 10 highest recorded 30-day commit counts and sit at star rank 50 or below. They are comparison cases, not a commit leaderboard.

Concentration here is the share of the cohort sum held by the repositories with the largest values of that metric. It is not a repository ranking and not a category mix.

Recorded 30-day commit volume is far more concentrated than accumulated stars: the top 10 hold 65.7% of recorded 30-day commit volume versus 14.0% of star mass. Forks sit in between — the top 20 hold 32.0% of cohort forks versus 23.6% of stars. That is the shape of this cohort's GitHub attention versus its recorded short-window commit volume.

63 of 326 eligible repositories (19.3%) are assigned to the AI Agents category — second only to LLMs (68). They hold 18.6% of cohort stars, roughly proportional to repository count. That is a cohort share in this snapshot's category assignment, not a ranking of agent products.

The category includes agent runtimes, orchestration, and adjacent tooling because assignment follows the snapshot's repository-classification rules rather than a manual product taxonomy. For a dated ranking of the AI Agents category, see Top AI Agent GitHub Repositories 2026.

Language is GitHub's primary-language field — a single Linguist label, so it describes the repository's primary detected language rather than its complete implementation stack. Missing values are counted as Unknown (13 repositories). Languages below 6 repositories are grouped as Other on the chart; the underlying counts are unchanged.

Python is the primary language for 49 of 61 repositories assigned to the PyTorch category (80.3%) and for 20 of 32 assigned to Transformers (62.5%). MCP Servers has the strongest TypeScript representation among the eight categories: 10 of 25 repositories (40.0%) list TypeScript as their primary language, versus 7 Python (28.0%). AI Agents is still Python-led (47.6%) with a larger TypeScript share (20.6%) than PyTorch.

Observed facts

• The eligible cohort on 17 September 2026 contains 326 repositories in eight operational categories.
• LLMs, AI Agents, and PyTorch together account for 192 repositories in this snapshot's assignment.
• 7 repositories have zero recorded stars in this snapshot.

Derived measurements

• Top 10 star share is 14.0%; top 10 recorded 30-day commit share is 65.7%. The two top-10 sets overlap in 3 repositories.
• 69.6% of repositories have a GitHub push within 30 days; 41.4% have zero recorded 30-day commits.
• AI Agents is 19.3% of the cohort by repository count and 18.6% by stars.

Cautious interpretation

• The snapshot is a tracked eligible AI/ML GitHub cohort with a wide range of repository ages and star totals under eight operational category assignments. It should not be interpreted as a map of all open-source AI, or of software quality.
• This should not be read as a forecast of which projects will matter next. The tracking window began 10 July 2026; the metrics are a point-in-time cut on 17 September 2026.

Dataset source: DataAIHub's tracked GitHub repository cohort. This article freezes the 17 September 2026 eligible repository snapshot and enriches each repository with the GitHub metrics described below.

Repositories enter tracking through multiple GitHub search and topic signals, followed by an AI relevance filter. Tracking began 10 July 2026. Category assignment uses discovery signals plus GitHub topic mapping.

The analysis uses repository category, primary language, stars, forks, contributors, last-push date, recorded 30-day commit volume, and repository creation date. Recorded 30-day commits are derived from GitHub's four-week commit-activity data. When that activity data is incomplete or unavailable, the value is treated as zero for this snapshot. A recorded zero is therefore a measurement result for this snapshot, not independent proof that a repository had no development activity. License is present for 276 of 326 repositories and is not used in the analyses above.

Seven repositories had zero recorded stars at snapshot time. They remained in the defined 326-repository cohort rather than being removed. They therefore contribute zeros to aggregate sums and can affect means, while having limited effect on medians and top-share concentration. No eligible repositories were dropped from the published cohort.

This page is a frozen research snapshot. Live daily rankings remain at GitHub Rankings. The August 2026 Top 20 mid-year article is a different cut of an earlier date and should not be mixed with these September figures.

• Top AI GitHub Repositories 2026 →
• Top AI Agent GitHub Repositories 2026 →
• Top Open-Source & Open-Weight AI Models 2026 →
• Live AI GitHub Rankings →
• GitHub Ranking Methodology →
• GitHub Trending →
• AI Agents on GitHub →
• LLM Repositories on GitHub →
• RAG Repositories on GitHub →
• AI Agents Guide →
• RAG Guide →
• Model Context Protocol →
• Agent Stack Landscape →
• Vector / RAG Stack Landscape →