GET-only

Check if self-hosted AI on a host you own is open

29 tools: Ollama, n8n, Langfuse, and the rest. Same unauthenticated GETs anyone can send. I never fetch coding-agent files on your box.

  • GET-only
  • only the host you submit
  • no login

Only scan assets you own or are authorized to test.

# no auth attempts. no exploits. results in ~30s. 102 paths from one decoy session. how it works

what an open Ollama answers
GET :11434/api/tags 200 OK {"models":[{"name":"llama3.1:70b"… GET :11434/api/version 200 OK {"version":"0.5.4"}
anyone on the internet can list, run, and copy your models.
what it should answer
GET :11434/api/tags connection refused (not public)
bound to localhost. the wall holds. this is the goal.

Why this is safe to run: methodology. GET-only, only the host you submit, nothing phones home.

see a full sample reportsimulated
case #2618A3DA · subject acme-ml-prod.example.com · probes 41 GETs
gradeF
Ollama · CRITICAL · model server, wide open
GET :11434/api/tags 200 OK {"models":[{"name":"llama3.1:70b"},{"name":"qwen2.5-coder:32b"}]}
n8n · HIGH · workflow automation, readable
GET :5678/rest/workflows 200 OK 14 workflow names · webhook URLs like /webhook/a3f9…c1
Langfuse · INFO · reachable, but the wall holds
GET :3000/api/public/health 401 traces stay behind the login · not graded

# SIMULATED. Same view a live scan of your host renders. The data here is fake. see the full attacker-eye view →

01

what I check

every cell = one GET probe
Ollama
GET :11434/api/tags
n8n
GET :5678/rest/settings
Open WebUI
GET :8080/api/config
vLLM
GET :8000/v1/models
Langfuse
GET :3000/api/public/health
ComfyUI
GET :8188/system_stats
Ray
GET :8265/api/jobs/
Qdrant
GET :6333/collections
Milvus
GET :9091/healthz
Dify
GET :5001/console/api/setup
AnythingLLM
GET :3001/api/ping
Jupyter
GET :8888/api/status
Gradio
GET :7860/config
Langflow
GET :7860/api/v1/version
Flowise
GET :3000/api/v1/ping
Chroma
GET :8000/api/v2/heartbeat
Weaviate
GET :8080/v1/meta
Redis
GET :5540/api/databases
LangServe
GET :8000/openapi.json
OpenClaw
GET :18789/__openclaw/control-ui-config.json
OpenHands
GET :8000/server_info
OpenAI-compatible API
GET :443/v1/models
AutoGen Studio
GET :8000/api/version
CrewAI Studio
GET :8000/auth/login
MCP
GET :3000/sse
MLflow
GET :5000/version
LiteLLM
GET :4000/health
Kubeflow
GET :8080/api/workgroup/env-info
LM Studio
GET :1234/api/v0/models
 → full coverage list

Run it yourself. Same engine, on your machine:

pip install aicheck-scan && aicheck agents --ci

Gate PRs so .claude / .codex files never land in git or a Docker context: aicheck agents.