Top GitHub Repositories in the AI Agents Ecosystem 2026

1. 1. Hermes Agent leads the agents category

   NousResearch/hermes-agent ranks #1 among repositories classified as agents with a DataAIHub Score of 89.61. It is also #2 in the full 326-repository eligible AI cohort, and it leads this agents Top 20 in stars (230,233) and recent commits (5,364).

2. 2. Agents are a large slice of the tracked AI cohort

   63 of 326 eligible repositories (19%) carry the agents category — second only to llm (67). This article ranks the top 20 of those 63 using stored DataAIHub Scores; it does not rescore the subset.

3. 3. DataAIHub Score is not a star ranking

   CopilotKit ranks #5 with 36,750 stars (16th by stars in this Top 20) alongside 1,663 recent commits and 1,429 releases. Shubhamsaboo/awesome-llm-apps has 132,539 stars (5th by stars here) but ranks #17, with 0 published releases and 43 recent commits.

4. 4. Recent activity is concentrated in the Top 20

   14 of 20 ranked repositories recorded at least 100 commits in the last ~30 days, and 5 recorded at least 500. All 20 had GitHub push activity within seven days of this snapshot.

5. 5. Frameworks sit beside adjacent platforms

   LangChain, AutoGPT, Langflow, CrewAI, LlamaIndex, and Haystack are agent runtimes or orchestration stacks. The same agents category also includes agent-adjacent infrastructure and applications — MLflow, Nx, SiYuan, ToolJet, career-ops, and a curated app list — because category assignment uses the existing seed/topic pipeline, not a manual re-label.

Rank is the order of repositories whose pipeline category is agents. The DataAIHub Score (0–100) was computed against the full 326-repository eligible AI cohort, then the agents subset was taken. Repository names link to the canonical GitHub URL from the ranking data.

Leaders below are calculated from this agents Top 20 only. Overall DataAIHub Scores were normalized against the full eligible AI cohort.

Profiles use verified GitHub metadata from the same snapshot. The type label is an editorial grouping from topics and descriptions; the ranking category for every row remains agents.

Every repository below is classified as agents in DataAIHub's GitHub ranking pipeline (seed query ai-agent plus topic mapping). The groups are an editorial reading aid from GitHub topics and descriptions — not a second official category system. Some groups (infrastructure, platforms, applications, curated lists) are agent-adjacent: they share the agents category but are not all agent runtimes.

Just outside this list, langchain-ai/langgraph ranks #22 among agents (score 69.16, 39.6K stars) in the same snapshot.

The DataAIHub Score combines log-normalized GitHub signals with fixed weights, scaled to 0–100 relative to the eligible AI cohort. This article does not introduce agents-only weights.

• Stars — Community interest and adoption signal on GitHub.
  35%
• Forks — Developer usage and extension signal.
  20%
• Commit activity — Recent development activity (last ~30 days via commit-activity stats).
  20%
• Contributors — Breadth of project participation (pipeline capped at 1,000).
  10%
• Watchers — Ongoing notification interest (GitHub subscribers_count when available).
  10%
• Releases — Published release activity on GitHub.
  5%

Full scoring detail is on the GitHub rankings methodology page. The live series is at GitHub rankings · agents.

DataAIHub Score ≠ GitHub stars. Stars are one adoption signal (35% of the score). Forks, recent commits, contributors, watchers, and releases also move rank. A repository with fewer stars is not automatically better — it can simply have stronger complementary signals in this cohort.

• NousResearch/hermes-agent leads both score (89.61) and stars (230.2K) in this agents Top 20.
• CopilotKit/CopilotKit is #5 on score with 36.8K stars, 1.7K recent commits, and 1.4K releases.
• Shubhamsaboo/awesome-llm-apps has 132.5K stars — more than CopilotKit — but ranks #17 with 0 releases and 43 recent commits.

In this snapshot the agents category is not only autonomous agent runtimes. It mixes frameworks (LangChain, CrewAI, LlamaIndex, Haystack), visual orchestration (Langflow, Flowise, Sim), a memory layer (Mem0), coding-adjacent frontend and monorepo tooling (CopilotKit, Nx), and application-specific agents (career-ops, daily_stock_analysis).

Tool use and orchestration show up in topics such as MCP, workflow automation, and multi-agent labels. LangGraph — a common orchestration substrate in DataAIHub's agent architecture and tool-calling guides — is tracked in the same category but sits just outside this Top 20.

12 of 20 ranked repositories are Python; the rest in this Top 20 are TypeScript or JavaScript. That matches a framework-and-workflow ecosystem more than a model-training one.

Use this research snapshot to enter the agents cluster. For the current table, use the live GitHub rankings; the guides and labs below cover agent design and practice.

Source: DataAIHub GitHub rankings API, category filter agents, snapshot 2026-08-14. Scores come from backend/app/services/repository_scoring.py as stored by the daily pipeline. Tracking in this dataset began 2026-07-10.

• Category membership follows the existing seed query and topic map. Some ranked repositories are agent-adjacent platforms, catalogs, or applications rather than agent runtimes. This article does not reclassify them.
• Scores are relative to the 326-repository eligible AI cohort, not an agents-only rescore.
• GitHub stars measure attention, not software quality.
• Contributor counts are capped at 1,000 by the ranking pipeline.
• This page is a frozen research snapshot of 2026-08-14; the live agents ranking can move the next day.

• Top AI GitHub Repositories 2026 →
• AI Agents on GitHub →
• Live AI GitHub Rankings →
• GitHub Ranking Methodology →
• AI Agents Guide →
• Agent Architectures →
• Tool Calling →
• Multi-Agent Systems →
• Agent Memory →
• Agent Loop Lab →
• Tool Calling Lab →
• Best AI Agent Frameworks →