How to Assess an AI Project on GitHub
7 Aug 2026 · RS Management
TL;DR
- A star is a bookmark almost nobody removes.
- A ranking can point to a project being wound down: Microsoft AutoGen, with over 60 thousand stars, is in maintenance mode, while its official successor has about 13 thousand.
- Six signals say more about risk than a ranking position and take fifteen minutes.
“Is this tool already the market standard?” In board meetings the answer usually comes down to one number: stars on GitHub. The number is public, easy to compare and looks objective, so it ends up in presentations and in architecture decisions. Before such a decision you need to know what that number actually measures.
The landscape in August 2026
The values below are star counts rounded to the nearest thousand.
MCP in the charts below stands for Model Context Protocol, the open standard for connecting AI agents to company tools.
Agent frameworks and platforms
- OpenClaw386k
- AutoGPT186k
- LangChain144k
- browser-use108k
- MCP servers89k
- OpenHands83k
- Cline66k
- CrewAI57k
- LlamaIndex51k
- LangGraph39k
The Model Context Protocol servers repository, visible in the first chart, has become the de facto exchange point for integrations between tools.
Running models on your own infrastructure
- Ollama178k
- Transformers163k
- llama.cpp123k
- vLLM88k
- LlamaFactory74k
- Unsloth70k
- LocalAI48k
- SGLang32k
RAG and semantic search
RAG (retrieval-augmented generation) is the pattern where a system first retrieves passages from company documents and only then attaches them to the prompt.
Workflow orchestration and interfaces
- n8n200k
- Langflow153k
- Dify152k
- Open WebUI148k
- AnythingLLM64k
- LiteLLM56k
- Flowise55k
- LibreChat42k
The charts fall into four layers that make up a typical AI solution in a company: an agent or platform that carries out tasks, an engine that runs the model, search across company documents and a tool that ties workflows together and gives users an interface. A project’s position within its layer says more than its position across all of GitHub, because it compares tools solving the same problem.
What popularity tells you
| Information | What it means for the company |
|---|---|
| Availability of skills | Easier to find a contractor, hire and replace someone who leaves the team |
| Integration surface | Ready connectors to systems the company already uses |
| Quality of materials | Documented bugs, deployment examples and AI models that know the project’s API (application programming interface) |
| Attention from security researchers | Vulnerabilities in widely used components are found faster |
For a team still building skills, the first two items weigh the most. A popular project shortens onboarding for a new person, because answers to typical problems are already written up and an AI coding assistant knows its functions from training data. With a niche project the same questions have to be solved alone, which raises the cost of every change.
What popularity does not tell you
Whether the project is still developed. A star can be removed with one click1, but the counter almost only ever grows, so a project being wound down can hold a high position for years. Microsoft AutoGen has over 60 thousand stars, and the latest code changes date from 15 April 2026.2 The repository README states that it is in maintenance mode and points to Microsoft Agent Framework, which has collected about 13 thousand stars. Choosing by ranking would mean picking the project being wound down instead of its official successor.
What the license is. n8n, with 200 thousand stars2, is released under the Sustainable Use License, and the project itself does not call itself open source, because the license restricts commercial use in a way the Open Source Initiative definition does not allow. Dify uses a modified Apache 2.0 license that forbids running multi-tenant environments without separate permission and removing the logo from the interface. For a company that wants to resell a service to end customers this difference is decisive and completely invisible in the star count.
What the risk profile is. OpenClaw earned its position as an assistant running on the user’s computer, executing system commands and file operations, controlled from messaging apps. That is a source of enthusiasm and of serious reservations wherever personal data or trade secrets are involved. The pace of star growth shows market curiosity, not readiness for corporate use.
Whether the project fits the task. A ranking does not know the document volume, response time requirements or that the documents are scanned PDFs in a language other than English. The statistically best project can be oversized or lack the one feature that decides everything in a given case.
Six signals instead of a ranking
No signal needs tools or access beyond the public repository, and together they take fifteen minutes. The first three tell you whether the project is still alive, the next three whether it is safe to build on.
| Signal | What to check | Warning sign |
|---|---|---|
| Last commit | The date of the last code change, not the last release | A fresh release but no commits for months |
| Release cadence | Whether releases come regularly | Three releases in a week, then six months of silence |
| Contributor concentration | Who makes the changes | Most changes from one account: the project depends on one person |
| Issue responses | Whether issues get a reply | 20 open issues and silence for six months |
| License history | Whether the license terms have changed | The project has tightened its terms once and may do so again |
| License text | The full text, not the label in the description | Restrictions on resale or multi-tenant environments |
Signals can conflict, such as fresh commits from a single active author. The layer where the component will sit, described in the next section, then decides. A longer assessment adds a security policy and the question of who funds development. A commercial owner usually means predictable releases, but also the risk of a license change once an investor starts expecting revenue.
Layers differ in the cost of a mistake
| Layer | Cost of replacement | Approach |
|---|---|---|
| Running models (Ollama, vLLM, llama.cpp) | Low: interfaces are similar and migration takes days | Room to experiment with younger projects |
| Agent and workflow orchestration (n8n, Dify, LangGraph) | High: it runs through business logic and team habits | A careful, proven choice |
The same weakness in a project means something different depending on how expensive it would be to back out of the decision. Open source components in an AI system are also subject to the same documentation regime as the rest of the solution, including the obligations under the EU AI Act. We assess specific tools and component choices as part of an AI Consultation.
Frequently asked questions
Does the number of GitHub stars show a project's quality?
Not directly. It shows interest and availability of skills, but not whether the project is developed, what the license is or whether it fits the task.
How do you check whether an open source project is safe to use in a company?
Check the date of the last commit, release regularity, the number of active contributors, issue responses, license history and the full license text. It takes about fifteen minutes.
Is n8n open source?
Not in the Open Source Initiative sense. n8n is released under the Sustainable Use License, which restricts some commercial uses.
Footnotes
-
GitHub Docs, “Saving repositories with stars”, the unstar step (“to unstar a previously starred repository, click Starred”): https://docs.github.com/en/get-started/exploring-projects-on-github/saving-repositories-with-stars. ↩
-
Public star counters on the repositories: OpenClaw, React, n8n, Dify, Microsoft AutoGen, Microsoft Agent Framework. ↩ ↩2
Blog content is informational and educational. It does not constitute legal or tax advice, nor individual business advisory. The scope of our services is described in the terms.
See the package: AI Consultation