Named after the AI from Star Trek Discovery. The one that merged with the ship and actually remembered everything. Built this after watching the OpenClaw security mess unfold. The core problem isn't OpenClaw specifically. Most agents run with everything permitted unless you restrict it. One misconfiguration and you've handed your filesystem to an LLM. Zora flips that. Zero access until you unlock it. Policy lives in policy.toml, loaded from disk before every action — not in the conversation where context compaction can erase it quietly mid-session. A few things worth knowing: No credit card required. Background tasks route to local Ollama by default, zero cost. If you want stronger models, it works with your existing Claude account via the agent SDK or Gemini through your Google account. Three memory tiers: within-session (fresh context injected at start), between-session (plain text files that survive restarts), and long-term consolidation (weekly background compaction scheduled at 3 am to avoid API cost). Memory survives. That's the point. Prompt injection defense via dual-LLM quarantine. Raw messages never reach the main agent. Happy to answer questions about the architecture.

Zora

Local-first AI agent. Zero trust. Memory that survives.

Local-first AI agent. Zero trust. Memory that survives.