Do one thing.
Do it well.
The Unix philosophy has guided software design for over 50 years. We believe it's the right foundation for building trustworthy, secure AI agents.
What is the Unix philosophy?
The Unix philosophy, originating from the design of the Unix operating system in the 1970s, is built on a simple principle: write programs that do one thing and do it well. Instead of building one massive, monolithic application, you build many small, focused tools that can be composed together.
This approach has proven remarkably resilient. It's the reason the command line is still powerful today, and it's the reason microservices architecture took over the cloud. Small, well-defined components are easier to understand, test, secure, and trust.
Single responsibility
Each program does one thing. If you need new functionality, build a new program rather than complicating an existing one.
Composability
Programs work together by passing structured data between them. The output of one becomes the input of another.
Clear boundaries
Each program has a well-defined interface. You know exactly what goes in and what comes out. No hidden side effects.
Unix philosophy in the app era
The smartphone era adopted Unix philosophy naturally. Every app on your phone is designed to do one thing and do it well: your camera app takes photos, your calculator calculates, your calendar manages events.
More importantly, each app comes with a capability manifest — the permissions dialog. When an app asks for access to your camera, location, or contacts, you see exactly what it needs. You can grant or deny each permission individually. This transparency has been the backbone of user privacy on mobile platforms.
This model worked because each app had a narrow scope. A camera app doesn't need your financial data. A calculator doesn't need your contacts. The single-purpose design made permission management intuitive and trustworthy.
Unix philosophy in the agentic era
The AI agent landscape today has mostly abandoned this principle. The dominant approach is to build one general-purpose agent that does everything — reads all your data, has access to all your tools, and tries to handle any request.
We think this is fundamentally wrong. Most of the time, the broad data access isn't actually needed, nor is access to all skills. An agent that drafts emails doesn't need to read your health records. An agent that manages your calendar doesn't need access to your financial data.
One agent with access to everything. If it malfunctions or is compromised, all your data is exposed.
Each agent has minimal, declared access. If one fails, the blast radius is contained.
Just like how the app permission model made mobile computing trustworthy, capability manifests make AI agents trustworthy. You always know exactly what each agent can and cannot do.
The Agent Store
The Agent Store is where you browse and select agents that fit your life. Each agent automates a specific aspect of your workflow, with clearly declared capabilities so you know exactly what you're getting.
The Unix philosophy makes adoption seamless: you understand what each agent does, what data it needs, and you can add or remove agents at any time without affecting others.
Email Drafter
Composes contextual email drafts based on conversation history and your tone.
Meeting Prep
Pulls relevant context before every meeting — attendee info, past notes, agendas.
Travel Planner
Detects upcoming trips and prepares logistics — hotels, weather, out-of-office.
Expense Tracker
Identifies receipts in your email and organizes them for reporting.
Follow-up Nudger
Monitors conversations that went cold and reminds you to follow up.
Research Assistant
Compiles background information on people and companies before key interactions.
Social Connector
Identifies warm introductions in your network for people you want to reach.
Deal Tracker
Monitors ongoing negotiations and surfaces key action items.