How to describe harperOS.
Positioning, architecture, approved language, and the vocabulary that keeps the product consistent — for partners, press, analysts, and the team.
The operating system for companies that run on AI.
harperOS is a single operating layer that holds a company's knowledge, runs the agents that act on it, and connects every department through one consistent system. It sits above the AI tools a company already uses and organizes them around the work, not the other way around.
Built on Claude Code as the core infrastructure. Tool- and model-agnostic by design.
Three connected layers.
Every harperOS deployment organizes around the same three layers. Each layer compounds the others.
The company's memory.
Curated, structured, and continuously updated. The knowledge layer is the company's accumulated expertise — products, customers, decisions, methodology, voice — held in a form agents can use.
The experts that act.
Each agent owns a defined function — sales analyst, account manager, marketing strategist, financial controller. Agents read the knowledge layer, execute work, and compound results back into the system.
The work itself.
Workflows orchestrate agents and knowledge into repeatable outcomes — from one-off tasks to entire cross-department operations. Every workflow gets better as the system runs.
Every agent is specialized.
harperOS agents are not generic copilots. Each one is built around five components that make it an expert in a specific function — and keep it sharpening over time.
Each agent reads from a defined slice of the knowledge layer — methodology, examples, voice, and prior decisions relevant to its function.
Repeatable workflows the agent runs. Skills are versioned, testable, and improvable. The agent gets sharper as the skills get sharper.
Scoped access to the systems the function needs — email, calendar, drive, CRM, finance, analytics. Tools are gated per agent; no tool is everywhere by default.
Every agent runs inside a system of rules and hooks that guarantee discipline — voice, evidence, source integrity, approval gates. Behavior is contracted, not improvised.
Every output is a signal. Corrections flow back into knowledge, gaps surface missing skills, and patterns harden into new rules. Agents compound — they get sharper with every run.
From decision to delivery.
Every harperOS deployment follows the same three-step path.
Decide on where to start.
Pick the first domain — the function that creates the highest leverage. Usually a department feeling the most pressure or sitting on the most underused knowledge.
Live OS in week one.
Within seven days, the operating system is live for the chosen domain — knowledge installed, agents deployed, workflows running.
Agents delivered.
Production agents work alongside the team from day one. Output per team member doubles. The system compounds as the next domain comes online.
Claude Code as the core infrastructure.
harperOS runs on Claude Code — Anthropic's agent infrastructure — as the execution backbone. Claude Code provides the runtime, tool surface, and discipline for every agent in the system.
The product is tool- and model-agnostic by design. Claude is the default model; the architecture supports the model and tool choices a company already runs.
Key terminology.
Words to use, and what they mean inside the system.
The full stack that holds knowledge, runs agents, and orchestrates workflows. Not a chatbot. Not a tool. A system.
The structured, curated, continuously updated record of how the company works — methodology, voice, decisions, examples.
A specialist that owns a defined function. Reads the knowledge layer, executes skills, and uses scoped tools.
A repeatable workflow an agent runs. Versioned, testable, and improvable. The unit of capability.
A bundle of agents, skills, and knowledge that serves a domain — sales, marketing, finance, account management, etc.
The orchestration of agents and skills into an end-to-end business outcome.
A functional area of the company — the unit at which harperOS is deployed.
The accumulation of decisions, patterns, and learnings the system captures and reuses. The reason output keeps getting better.
Anti-positioning.
Easier to understand harperOS by what it is not.
Not a chatbot.
harperOS does not sit behind a prompt box waiting for questions. Agents are scheduled, triggered, and workflow-bound.
Not a model wrapper.
Models are interchangeable. The operating system is what compounds — knowledge, skills, rules, and the workflows that connect them.
Not a vertical SaaS.
harperOS is horizontal — it runs every department inside one operating layer, not a single function inside one application.
Not a pilot.
Built inside live operations from day one. Deployed against real work, with real teams, in production. No sandbox phase.
How to describe it.
Phrases approved for partner, press, and analyst use.
harperOS. One word, lowercase “h,” uppercase “OS.” The wordmark uses an underscore between “harper” and “OS” as a signature element — apply the underscore in display, not in body copy.
Enterprise AI infrastructure. The operating system for companies that run on AI.
harperOS is the operating system for companies that run on AI.
People command. Proxies execute. Knowledge compounds.
For deeper reference.
Talk to us.
For product questions, partnership inquiries, or further detail. Email us at pa@harperos.ai or press@harperos.ai