harness.os

An operating system for AI-assisted work

AI companies sell you a brainstem and a thin layer of memory — but with no cognitive structure and no way to carry it across agents. harness.os is the full cortex: structured knowledge that compounds, learns, and works with any agent. Your knowledge, your OS.

See the full vision Read the kernel docs
Scroll

Every OS concept maps 1:1.
Then comes the cognitive layer.

harness.os is a real operating system — not a metaphor. Processes, memory management, filesystems, system calls, drivers, boot sequences, security. Every concept from an OS course has a concrete implementation. "Everything is a file" becomes "everything is knowledge."

But then the mapping table keeps growing — and Linux stops. Learnings, decisions with rationale, metacognition, cross-context transfer. No OS has ever had these, because programs don't think. AI agents do. That cognitive layer is the cortex — and nobody else builds it.

Three layers — just like a real OS

Same architecture as Linux: kernel, distribution, runtime.

01 Kernel
The invariant core. Four harness types, five cross-cutting concerns, knowledge primitives, system calls (MCP tools). These structures apply to every installation — they are the "everything is knowledge" equivalent of Unix's "everything is a file."
Like the Linux kernel — processes, memory, filesystems, syscalls, security.
02 Distribution
Your specific installation. Subtypes (Build:Software, Product:SaaS), default rules, installed knowledge, preferred surfaces, sync policies. "marco.os" and "acme.os" are distributions — same kernel, different configurations. Multi-machine sync keeps cognitive data flowing between personal and work installations.
Like Ubuntu vs macOS — same kernel primitives, your choices on top.
03 Mesh
The running instances. build.ai, cortex.ai, way2do — each mesh connects harnesses at runtime, enabling cross-domain reasoning. Any agent (Claude, GPT, Gemini) can plug in as a process — the OS and its cognitive layer are what persist.
Like running processes — live services that start, work, and hand off context.

Four harness types

Four kinds of knowledge, each with its own behavior in the kernel.

Build
How you make things
Your creation process. Software development, content production, marketing — each a subtype with its own rules, workflows, and natural access surface.
Product
Why you build and what you ship
Why you build and what you ship. SaaS products, mobile apps, open-source projects — each subtype tracks its own lifecycle, roadmap, and session handoffs.
Operations
How your domain works
How your domain works. Skydiving progression, tax rules, legal compliance — each industry gets its own operations subtype with stable, reusable knowledge.
Domain
Who your users are
Per-user data within those domains. Jump logs, workout history, transactions. The stuff that makes responses specific to a person, not generic.

The mesh in action

Here's what happens when you connect the harnesses together.

"Can I afford this skydive camp next month?"
ops:skydiving
Camp Details
Dates, cost, requirements
ops:finance
Budget Rules
Income, expenses, targets
domain:personal
Your Data
Calendar, jump log, accounts
Mesh Response
The camp costs $1,200 and requires B-license (you have it). After rent and subscriptions, you'll have $1,450 free next month. You can afford it with $250 remaining.

Five cross-cutting concerns

Five things I kept needing to track across all four types.

Build
Product
Operations
Domain
Relational
How entities connect across domains. A skydive camp is also a budget item is also a calendar event.
Governance
Data quality, audit trails, compliance, validation rules.
Causal
The WHY behind decisions. Without this, every new session re-debates settled questions.
Metacognitive
Knowledge about knowledge. What's stale, what works, what needs updating.
Security
Threat models, risk assessment, data protection, access control.

Agents are plug-ins. Knowledge is the OS.

The outer harness holds all intelligence. The inner harness connects any agent — Claude, GPT, Gemini — to the same OS. Swap the agent; the knowledge, learnings, and decisions persist.

Outer Harness (substance)
The OS itself — kernel + cognitive layer. Knowledge, rules, workflows, learnings, decisions. This persists across agents, sessions, and machines.
  • Knowledge, rules, workflows
  • Decisions and learnings
  • Governance and audit trails
  • Mesh topology
  • Cross-domain relationships
Boundary
Inner Harness (surface)
The agent connector. MCP protocol, tool definitions, and an adapter per agent (CLAUDE.md, .cursorrules). Any model becomes a process on the OS.
  • Claude Code — via MCP tools, like a terminal session
  • GPT / Gemini — same MCP protocol, same knowledge
  • Files — natural for Build:Software, syncs with MCP
Key insight: There shouldn't be "make your agent smarter at X." There should be: define the knowledge and workflows your OS needs — then let any agent execute them. If your knowledge lives in your OS with your cortex, their model becomes a plug-in.

From files to an OS — in 11 weeks

It started with CLAUDE.md files across six products. Those became structured rules, then workflows, then a full knowledge system. When I studied the OS fundamentals I'd missed in college, every concept mapped 1:1 to what I'd already built. It wasn't a methodology — it was a real operating system. And on top of it, a cognitive layer that no OS has ever had.

The vision page has the full story — from computing history to why this layer was inevitable.

Read the vision Read the blog