nightluna.com

What It Is

Luna AI System is Luna’s internal AI workplace, not a general-purpose AI SaaS product sold to the public. It brings together company employees, Claude-based agents, tools, workflows, and company knowledge entry points into an “operating system for how Luna works.” Its core principle is “People lead; technology amplifies”: humans remain responsible for judgment and approvals, while AI handles repetitive work, organization, and drafting.

Core Capabilities and Use Cases

Based on the main text, the platform includes several major modules: Agents, Tools, Workflows, Fleet, and Company/governance documentation. Its agents include Basal, Pen, Clara, Pump, Sensor, and others, covering shared assistants for all staff, personal owner agents, and internal platform agents. Typical tasks include company Q&A, document drafting, summarization, brainstorming, code snippets, CI failure summaries, first drafts of tests, support ticket triage, morning briefings, fleet briefings, and assistance with low-judgment steps in clinical and regulatory materials. It also provides roughly 20-minute onboarding tracks tailored to engineering, clinical, operations, leadership, and admin roles, with an emphasis on what must not be delegated to AI.

Models, Integrations, and Governance

The text explicitly states that the agents are built on Claude, using infrastructure such as Cloudflare Workers, Durable Objects, D1, R2, AI Gateway, Pages Functions, and GitHub Actions. For integrations, it emphasizes a Slack-first approach, allowing employees to interact with agents within their existing work context. Governance is one of the system’s more notable aspects: every agent/workflow has an owner, a “why,” and a definition of what it can’t do. Basal chat already includes a kill-switch and audit-log. Audit records capture call metadata and hashes, but not the prompt or response content.

Privacy, Data Boundaries, and Limitations

The privacy design is disclosed in relatively fine detail: conversations are private and not used for model training; Anthropic Zero Data Retention is contractually enabled; each user has a separate Durable Object for storing memory; users can view and delete memory under My Agents; and PHI remains within HIPAA boundaries. The limitations are also clearly stated: agent outputs cannot replace human judgment, and anything related to regulatory matters, patients, clinical content, or investors must be reviewed by a human. Many capabilities are still on the roadmap, such as the briefing pipeline, workflow runtime, chat persistence, and a memory-clear runner.

Pricing, Target Users, and Access from China

The main text contains no public pricing, free trial, payment method, or external registration information, so it should be regarded as an internal Luna system. It is mainly useful as a reference for teams looking to build an internal enterprise AI agent operating system, especially medical, hardware, and operations-heavy organizations that care about compliance, auditing, and clear human-AI responsibility boundaries. Access from China is unknown. For purchasable alternatives, consider comparing it with ChatGPT Enterprise, Claude for Work, Microsoft Copilot, or self-hosted/semi-self-hosted agent platforms such as Dify and Coze.