{
  "slug": "ollama",
  "repo": "ollama/ollama",
  "name": "Ollama",
  "tagline": "Run, customize, and orchestrate large language models locally on your own hardware.",
  "category": "ai-agents",
  "license_spdx": "MIT",
  "stars": 186063,
  "contributors": 390,
  "last_push_days": 0,
  "latest_release": "v0.5.11",
  "safety_score": 93.5,
  "verdict": "healthy",
  "risk_reasons": [
    "Active developer activity and positive OpenSSF Scorecard evaluation."
  ],
  "scorecard": 8.8,
  "components": {
    "security_health": 94,
    "maintenance": 95,
    "community": 93,
    "releases": 90
  },
  "language": "Go",
  "self_host_difficulty": "Easy",
  "install_commands": {
    "curl": "curl -fsSL https://ollama.com/install.sh | sh",
    "docker": "docker run -d -v ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama"
  },
  "ai_report": "Ollama is the premier open-source local LLM runner and model manager, engineered in Go with high-performance llama.cpp C/C++ backend bindings. It abstracts model quantization (GGUF), GPU layer offloading (CUDA, ROCm, Metal), and prompt templating behind a clean, OpenAI-compatible REST API. By packaging weights, prompts, and system configurations into unified 'Modelfiles', Ollama makes pulling and running frontier open models like Llama 3.3, Mistral, Qwen 2.5, and DeepSeek-R1 as simple as a single CLI command.\n\nFrom a security and data sovereignty perspective, Ollama operates 100% offline once model weights are pulled. No prompt data, system prompts, embeddings, or inference outputs leave the local workstation or private server cluster. By default, Ollama binds its HTTP daemon strictly to loopback (127.0.0.1:11434), preventing unauthorized network discovery. In 2024, maintainers swiftly addressed CVE-2024-37032 ('Probllama' path traversal in API endpoints) and enforced strict validation routines for model blob paths.\n\nThe repository boasts an 8.8 OpenSSF Scorecard rating with automated CI testing, reproducible build scripts, and disciplined dependency updates. For organizations and developers demanding complete confidentiality for AI workloads, Ollama represents an essential foundation, achieving an outstanding Healthy rating of 93/100.",
  "scanned_at": "2026-11-06T08:58:06.864Z",
  "website_url": "https://ollama.com",
  "use_cases": [
    "Running state-of-the-art open weights LLMs (Llama, Mistral, Qwen, DeepSeek) locally without cloud subscription fees.",
    "Serving local OpenAI-compatible API endpoints for developer agents, coding assistants, and internal apps.",
    "Private document summarization and embedding generation with zero risk of proprietary data leaks."
  ],
  "requirements": {
    "cpu": "4 Cores (NVIDIA CUDA or Apple Silicon GPU strongly recommended)",
    "ram": "8 GB (16 GB+ recommended for 7B/14B models)",
    "disk": "20 GB",
    "runtime": [
      "Go",
      "macOS, Linux, or Windows"
    ],
    "difficulty": "Beginner-friendly"
  },
  "audience": {
    "skipIf": [
      "Users on extremely constrained hardware (under 4 GB RAM) without discrete GPU acceleration.",
      "Teams seeking turnkey visual chat web interfaces out-of-the-box without pairing with a web frontend like Open-WebUI."
    ],
    "perfectFor": [
      "Developers and homelabbers seeking an effortless, privacy-preserving way to run LLMs locally.",
      "Security-conscious organizations needing OpenAI-compatible API backends within air-gapped or private networks."
    ]
  },
  "cves": [
    {
      "id": "CVE-2024-37032",
      "url": "https://nvd.nist.gov/vuln/detail/CVE-2024-37032",
      "summary": "Path traversal vulnerability in Ollama API model pull handler leading to arbitrary file write.",
      "fixed_in": "v0.1.34",
      "severity": "high",
      "fix_status": "fixed",
      "published_at": "2024-06-19T00:00:00Z"
    }
  ],
  "permission_model": {
    "verdict": "Healthy: High-security model runner. Zero telemetry, local-only inference execution, and protected storage partitions.",
    "review_status": "approved",
    "network_access": "Loopback Only (127.0.0.1:11434 by default)",
    "filesystem_access": "Scoped Model Storage (~/.ollama)",
    "execution_environment": "Local Host Daemon / GPU Accelerated",
    "approval_prompt_detail": "Model execution runs in isolated daemon context; API endpoints serve read-only inference without direct host command execution.",
    "asks_before_destructive": true
  },
  "incident_history": [
    {
      "id": "CVE-2024-37032",
      "date": "2024-06-19",
      "link": "https://nvd.nist.gov/vuln/detail/CVE-2024-37032",
      "type": "tool-abuse",
      "title": "Ollama API Path Traversal Patched",
      "severity": "high",
      "mitigation": "Resolved in v0.1.34 by enforcing canonical path resolution and strict digest character verification.",
      "description": "Crafted digest parameters in model pull manifests allowed file writes outside the intended models cache directory."
    }
  ],
  "ai_report_status": "approved",
  "created_at": "2026-10-01T10:00:00.000Z",
  "updated_at": "2026-10-01T10:00:00.000Z",
  "archived": false,
  "unlisted": false,
  "setup_steps": {
    "status": "none",
    "steps": [],
    "source_url": "https://github.com/ollama/ollama#readme",
    "model": "ag/gemini-3.8-flash-high",
    "written_at": "2026-11-06T11:45:28.222Z",
    "reason": "no self-host instructions in the README"
  }
}