Research Loop is designed with a local-first architecture:

• All session data stays on your machine. Nothing is sent to any server unless you explicitly export a bundle or publish to the registry.
• LLM API calls transmit paper text and code snippets to your configured provider (Anthropic, OpenAI, Ollama, etc.). If you require full on-device privacy, use Ollama or LM Studio as your backend.
• API keys are read from environment variables only. Never stored in plain-text config files or committed to git.
• Paper library stores PDFs and extracted text locally. No institutional credentials or paywalled content is accessed automatically.
• .research bundles contain paper metadata but not full PDFs. Review bundles before public sharing if working on proprietary code or unpublished results.

Research Loop is a local developer tool, not a multi-user server. The primary risks are:

1. Accidental secret exposure — API keys or proprietary code in exported bundles or shared JSONL logs.
2. Malicious .research bundles — A bundle from an untrusted source could contain a benchmark_command that executes arbitrary shell commands. Only open bundles from sources you trust.
3. LLM prompt injection — A malicious paper could attempt to inject instructions into the LLM pipeline. The system does not have write access to your broader filesystem by default, but exercise caution with untrusted PDFs.

If you discover a security vulnerability, please do not open a public GitHub issue.

Instead, email: security@research-loop.dev

Include:

• A description of the vulnerability
• Steps to reproduce
• Potential impact
• Any suggested mitigations

You will receive acknowledgment within 48 hours and a resolution timeline within 7 days. We follow responsible disclosure — we ask that you give us 90 days to address the issue before public disclosure.

• Pin your config.toml to a specific model version to avoid unexpected behavior from model updates.
• Review auto-generated hypothesis.md before running experiments — the Empirical Agent executes the benchmark command as written.
• Do not run Research Loop with elevated privileges (sudo/root).
• Use .gitignore to exclude .research-loop/credentials.toml and any local PDF files from version control.