What is DOS

DOS is the Durante Operating System — a structured layer on top of Claude Code for repeatable work, memory, and capability orchestration.

DOS is the operating system you install on top of Claude Code so your most important work stops being improvised.

One sentence

DOS (Durante Operating System) is a structured layer that sits on top of Claude Code and adds three things Claude Code alone does not: a reproducible way to plan and execute complex work (the Algorithm), persistent memory that survives sessions (MemPalace), and a library of bundled capabilities you invoke by outcome instead of by name (Packs).

The stack

  • Claude Code is the model — raw intelligence, tools, the harness running on your machine.
  • DOS is the operating system — conventions, workflows, memory, and capabilities on top.

An operating system is not the CPU. It is what turns a CPU into a machine you can work on — file systems, process management, a predictable way to install software. DOS does the same for a language model: structure to start a task, run it to completion, verify it, and remember what happened.

What DOS adds

Four additions

The Algorithm, Packs, persistent memory, and execution modes. Everything else is a consequence of these four.

  1. The Algorithm — a 7-phase execution loop (OBSERVE, THINK, PLAN, BUILD, EXECUTE, VERIFY, LEARN) that kicks in automatically for non-trivial work. It forces reverse engineering of the request, atomic verifiable criteria (ISC), and explicit capability selection before any code is written.
  2. Packs — 26 bundled capabilities in v0.0.3 covering research, media, scraping, brand, sales, memory, security, and more. The Algorithm picks them based on the outcome you ask for.
  3. Persistent memory — MemPalace captures learnings, decisions, ratings, and TELOS (goals, beliefs, projects) across sessions. Semantic search over your own history becomes a tool call.
  4. Modes — NATIVE for quick tasks, ALGORITHM for complex work, MINIMAL for acknowledgments. Overhead matches the task.

One example

You say: "Give me a one-page brief on how post-quantum crypto affects our auth stack over the next 18 months."

Without DOS, you negotiate scope and formatting with the model. With DOS, the Algorithm enters OBSERVE, picks the Research Pack, generates atomic ISC ("brief fits one page", "covers NIST PQC finalists", "names two threats to current auth"), runs the research, verifies each criterion, and writes the brief. Come back tomorrow — MemPalace remembers the decision.

What DOS is not

  • A chatbot wrapper — no custom UI, no proprietary chat window, no hosted history. You use Claude Code the way you already do.
  • An autonomous agent platform — DOS does not run unsupervised in the cloud. Every Algorithm run happens inside your Claude Code session, with you in the loop.
  • A Claude replacement — DOS ships no model of its own. Every inference goes through the model Claude Code is running.
  • A framework you import into your codebase — DOS lives in your home directory. Your project stays yours.

If something sounds like magic, it is Claude Code doing the work and DOS giving it structure.

Who it is for

Developers who already use Claude Code seriously and have started to notice the same problems:

  • You re-explain the same context at the start of every session.
  • You reinvent the same workflow each time you do research, write a PRD, or ship a feature.
  • You have no memory of what you decided six weeks ago, and neither does the model.
  • You want to invoke specialized capabilities (deep research, scraping, brand, media) without stitching agents together by hand.

DOS does not replace the way you work. It stabilizes the parts of your work that keep slipping.

The shortest path in

Install DOS, open a Claude Code session inside a real project, describe an outcome. The Algorithm takes it from there. You do not need to learn Pack names, Skill names, or phases on day one. Go deeper when you want — by then DOS has remembered what you care about.

Was this page helpful?