Local AI vLLM
Is vLLM safe?
Short answer: Caution. StackGrade's weekly scan of vllm-project/vllm gives it a Caution label for a skill or agent instruction file (AGENTS.md) that tells the agent to pipe a remote script into a shell. It does list 1 high-severity advisory in runtime dependencies, which doesn't change the label. Automated scan of commit 0e5b2ba on 2 Nov 2026, not a manual audit.
Updated : the last week its score, rank or safety result changed.
High-throughput, memory-efficient engine for serving LLMs.
- 96,100 stars
- pushed 2 Nov
- Apache-2.0
- Python
#24of 52 overall#5in Local AI▼ 1 vs Week 44
- Maintenance 30%64
- Adoption 30%70
- Docs 20%13
- Community 20%–
Score over the last 4 weeks: 57, 55, 55, 54.
Skill/agent file instructs piping a remote script into a shell: AGENTS.md.
How this score works
Each part is a percentile among all 52 tracked tools, so 50 is average for this list. Community is not measured yet; the score uses the other 80%.
- Maintenance: Last push 2 Nov 2026. Last 10 GitHub releases within 17 days (about 16 per 30 days), latest release 1 Nov 2026. Issue close time: fewer than 5 closed issues, so not used.
- Adoption: 96,100 stars (70%), 24,281 forks (30%).
- Docs: README 5,770 characters, 6 headings, 2 code blocks, install section: yes.
#5 of 5 in Local AI
Ranked by score. The yellow row is this tool.
- 1Ollama#2 overall73Caution
- 2LobeHub#7 overall69Caution
- 3Open WebUI#8 overall68Caution
- 4llama.cpp#9 overall68Caution
- 5vLLM#24 overall54Caution
Safety scan, 2 Nov 2026
Automated scan of commit 0e5b2ba, 8,030 files. Not a manual audit.
Caution
- 0Critical dependency advisories1 high, 1 medium, 3 low. OSV, runtime dependencies only.
- 0Install hooksNo npm install scripts found.
- 0Remote-code patternsFor example eval(fetch(…)).
- 31Agent files checked1 file tells the agent to pipe a script into a shell.
- 5curl | sh installersIn docs or program code. Information only.
Why this label?
Why this label:
- Skill/agent file instructs piping a remote script into a shell: AGENTS.md — curl -LsSf https://astral.sh/uv/install.sh | sh
Known vulnerabilities in runtime dependencies (OSV): 0 critical, 1 high, 1 medium, 3 low, 31 unrated
Also found (informational, does not change the label):
- 1 high-severity known vulnerability in runtime dependencies (OSV)
- Program code runs a curl|sh style installer (3 occurrences, e.g. .buildkite/scripts/upload-release-wheels-pypi.sh)
- Documents curl|sh style installer (2 occurrences, e.g. docker/Dockerfile)
Notes:
- 4 affected packages only in dev/test/docs manifests, not counted
- Telemetry/analytics references in code: telemetry, amplitude
Automated scan of commit 0e5b2ba on 2 Nov 2026, not a manual audit. 8,030 files checked. What we check
Repo activity
From the GitHub API, fetched 2 Nov 2026.
- 10+Releases, 90 daysLatest v0.33.0, 1 Nov
- –Median issue closeFewer than 5 closed issues
- 96.1kStars+569 since 26 Oct
- 24.3kForks9,734 open issues
- 2023Repo created9 Feb 2023
10 releases in the last 30 days (GitHub releases feed, last 10 entries):
Questions about vLLM
Answers use this week's StackGrade data.
Is vLLM safe to use?
Caution. StackGrade's weekly scan of vllm-project/vllm gives it a Caution label for a skill or agent instruction file (AGENTS.md) that tells the agent to pipe a remote script into a shell. It does list 1 high-severity advisory in runtime dependencies, which doesn't change the label. Automated scan of commit 0e5b2ba on 2 Nov 2026, not a manual audit.
How does StackGrade score vLLM?
vLLM scores 54 out of 100 in Week 45: #24 of 52 tracked tools overall and #5 in Local AI. The score combines maintenance (64), adoption (70) and docs (13) as percentiles among all 52 tracked tools; community (20% weight) is not measured yet, so the score uses the other 80%. It is not a hands-on test and it is separate from the safety label.
What are the alternatives to vLLM?
The highest-scored other tools in Local AI this week: Ollama (73, Caution), LobeHub (69, Caution), Open WebUI (68, Caution), llama.cpp (68, Caution). StackGrade has not tested them hands-on.
What licence does vLLM use?
The licence listed for vllm-project/vllm is Apache-2.0. Read the licence text before commercial use.
Compare and use it
Side by side
Used in 1 AI use case
All facts and sources
- Type
- AI tool
- License
- Apache-2.0
- Language
- Python
- Created
- 9 Feb 2023
- Owner
- vllm-project (Organization)
- Forks
- 24,281
- Homepage
- vllm.ai
GitHub description: “A high-throughput and memory-efficient inference and serving engine for LLMs”
Sources: github.com/vllm-project/vllm, api.github.com/repos/vllm-project/vllm, raw.githubusercontent.com/vllm-project/vllm/main/README.md. Fetched 2 Nov 2026.
Know when vLLM changes label
One email a week: the new top 10, the biggest score changes and every safety label that changed.
Free. Double opt-in: you confirm by email first, unsubscribe in one click.