Claude Code is the engine. A framework is the car.
Out of the box, Claude Code is a powerful agent: it reads your files, runs commands, and writes code. But it forgets between sessions, re-derives the same workflows every time, and uses one model for everything — the cheap question and the hard one alike.
A Claude Code framework is the system that closes those gaps. It adds operator-owned memory so context compounds instead of resetting, a set of agent skills so your workflows are named and reusable, and a routing harness so each task runs on the cheapest model that can do the job. You still bring your own Claude — the framework is everything around it.
Bare Claude Code vs. Claude Code + a framework
| Bare Claude Code | Claude Code + Libro (framework) | |
|---|---|---|
| Memory | Forgets between sessions; you re-paste context every time. | Persistent local memory — decisions and notes compound across sessions. |
| Workflows | Re-explained from scratch each run as raw prompts. | Reusable agent skills — named, versioned, invoked on demand. |
| Model use | One model for everything, cheap task or hard. | Routing by cost and fit — local for cheap, cloud for hard. |
| Onboarding | Cold start every session; no shared house rules. | Session rituals brief the agent on open loops automatically. |
| Ownership | Whatever you wire by hand, undocumented. | Every file inspectable — Apache 2.0, runs on your machine. |
Three things a framework buys you
Context that compounds
Operator-owned memory means every session makes the next one smarter. You stop re-explaining your project on every open.
Skills, not prompts
A workflow you ran once becomes a skill you invoke forever — session-start, SEO audit, release. Reuse beats re-deriving.
Spend control
Routing sends the cheap question to a local model and saves the cloud bill for the work that needs it.
Libro is a Claude Code framework you own
Libro is a free, open-source framework for Claude Code. It ships operator-owned memory, a coordinated library of agent skills, and a model-routing harness — the whole layer above the agent, ready to install. You bring your own Claude subscription; Libro never proxies your traffic. Apache 2.0, runs on your machine, no servers.
See the framework
Free and open-source under Apache 2.0. Read the install script before you run it — that's the point.
Explore Libro →Related: Own your AI vs. rent it →
Claude Code framework FAQ
What is a Claude Code framework?
A layer of memory, skills, and model routing that sits on top of Claude Code. Claude Code is the agent in your terminal; a framework gives it persistent memory across sessions, a library of reusable skills, and rules for which model answers which task. Libro is one such framework — free, open-source, and operator-owned.
What are Claude Code skills?
Reusable, named workflows the agent can invoke on demand — a session-start ritual, an SEO audit, a release process. A skill packages the instructions, scripts, and context for one job so you don't re-explain it every time. A framework ships a coordinated set of skills instead of one-off prompts.
Is Libro free?
Yes — free and open-source under Apache 2.0. You read the install script before you run it, and there are no servers or subscriptions on Libro's side. You bring your own Claude Code subscription.
Do I need my own Claude subscription?
Yes. A framework like Libro is the system around the model, not the model itself. You bring your own Claude (or a local model via Ollama); the framework adds memory, skills, and routing. Libro never proxies your Claude traffic.
How is a framework different from bare Claude Code?
Bare Claude Code forgets between sessions, re-derives the same workflows, and uses one model for everything. A framework adds operator-owned memory so context compounds, a skill library so workflows are reusable, and routing so each task runs on the cheapest model that works.