davisxai/operator-knowledge-base

操作员知识库

真实的人工智能系统，有文档记录。OperatorOS背后的架构、指南、技能和工具。

operatoros.ai · @davis.mcm · @os.操作员

---

这是什么

我在社交平台上发布的所有内容都追溯到我实际运行的系统。这个仓库是这些系统被妥善记录的地方：免费、无限制，并且在设计决策层面而非截图。

我是戴维斯·麦克默雷恩。我运营着OperatorOS，一家人工智能系统机构。我为真实企业构建仪表盘、人工智能操作系统和自托管软件栈，我自己也使用同样的架构。内容展示了构建。这个知识库包含了你自己构建的实质内容。

这里每页都有三项承诺：

1. 它存在于生产环境中。 没有文档说明任何不实际运行的内容。失败模式被包含在内，因为学习就在那儿。
2. 没有虚构数字。 成本、计数和基准来自真实系统和真实研究通行证，关键处有日期。
3. 没有限制。 没有邮件墙，没有课程推销。给仓库加星号，新内容会在上传前到达这里。

[!注]

Operator Knowledge Base

Real AI systems, documented. The architecture, the guides, the skills, and the tools behind OperatorOS.

operatoros.ai · @davis.mcm · @os.operator

---

What this is

Everything I ship on social traces back to systems I actually run. This repo is where those systems get documented properly: free, ungated, and at the level of design decisions rather than screenshots.

I’m Davis McMurrain. I run OperatorOS, an AI systems agency. I build dashboards, AI operating systems, and self-hosted software stacks for real businesses, and I use the same architecture for my own operation. The content shows the builds. This knowledge base holds the substance so you can build your own.

Three commitments run through every page here:

1. It exists in production. Nothing is documented that does not actually run. Failure modes are included, because that is where the learning is.
2. No invented numbers. Costs, counts, and benchmarks come from real systems and real research passes, dated where it matters.
3. Nothing gated. No email wall, no course upsell. Star the repo and new material lands here before it hits the feed.

[!NOTE] Client names, contract values, credentials, and server addresses are deliberately excluded everywhere. The architecture is the interesting part anyway.

---

The map

ai-os/ The AI operating system

The system I actually run every day, not a rendered HTML page pretending to be Jarvis. Mail triage across three Gmail accounts, an agent-maintained knowledge vault, a chat agent with propose-only permissions, and a generated morning brief.

• architecture.md The runtime. Two processes that only talk through Postgres, the actions table as an approval gate, Gmail triage design, model tiering, cron cadences, and the security posture.
• vault.md The agent-maintained Obsidian second brain. Split by mutability, the index/hot/log triad, frontmatter as a validated contract, and how autonomous enrichment stays bounded.
• claude-code-layer.md Claude Code configured as a business operating system. The CLAUDE.md constitution, a 66-skill library, agent tiering, hooks, MCP lineup, and the memory pattern.