FAQ

FAQ

General

• Q: What is Zeref OS in one sentence?
  A: A persistent, harness-agnostic context and memory engine for AI projects, stored in local-first markdown files with privacy on by default.

• Q: Is this an agent or a standalone program?
  A: No – Zeref OS isn’t a running service. It’s a framework that wraps around existing LLM tools (Claude, Cursor, etc.) to give them persistent memory. It relies on those tools to execute the prompts defined by Zeref.

• Q: Is Zeref OS open source?
  A: Yes, it’s MIT-licensed (see LICENSE), and free to install. You bring your own LLM models; we don’t provide any in the repo.

• Q: Does it send my data to the cloud or OpenAI?
  A: No. By default, all data is local. External tool calls or sharing require explicit user setup (via SHARING_POLICY.md). Even if you connect an LLM API, Zeref itself does not phone home.

Setup

• Q: How long does setup take?
  A: About 5 minutes. Running /zeref-os:start conducts an interactive setup, creating 7 configuration files (project name, budget, privacy mode, etc.) and initializing memory/. The CLI commands (/start, /stop) are the main interface.

• Q: What if I stop the interview midway?
  A: Zeref OS will boot in read-only mode until the missing parts are completed (per the spec in MIGRATION.md). You can restart /zeref-os:start to resume.

• Q: Do I need any special tools or subscriptions?
  A: No. Zeref OS itself has no paid components. You do need access to an LLM (e.g. an API key or a local model) in your chosen harness.

Privacy

• Q: What is the default privacy mode?
  A: abstract. In this mode, Zeref REDACT.md rules will scrub PII from outputs, replacing details with high-level descriptions.

• Q: Can I ever store exact secret data?
  A: Yes. Setting mode: exact in the root PRIVACY.md disables abstraction. Use this if you explicitly want to log an exact password or email. Use with caution.

• Q: How do I ensure nothing ever leaves my project?
  A: Set mode: local-only in PRIVACY.md. In this mode, all external API calls and syncs are blocked. Any attempt to write to memory/sync/outbound/ or push to a parent will be denied.

• Q: Why are there three files (PRIVACY, REDACT, SHARING_POLICY)?

  • PRIVACY.md = global mode and policy (abstract/exact/local-only).
  • REDACT.md = specific tokens or classes to sanitize.
  • SHARING_POLICY.md = which connectors (e.g. Slack, GitHub) are allowed to see data (defaults to none).