Starizzi Desktop: the AI agent operation hub of izziapi.com
In the previous seven articles, we learned how to think about an agent: memory, skills, control, delegation, distribution, and connection. However, these concepts need a place to live and run — a cockpit where you can actually install agents, assign tasks, and see them collaborate.
With the izziapi.com ecosystem, that cockpit is Starizzi Desktop (build name: izzi‑OpenClaw) — a desktop application for installing, running, and organizing AI agents, directly connecting to the Izzi platform. This article analyzes Starizzi from an architectural perspective: what it is, what components it includes, and why it embodies the toolkit we have learned.
Note: Starizzi is rapidly evolving; this article describes it at a level of architecture and philosophy (stable parts). For feature names, lists of pre-built agents, and specific download links, you should refer to the latest release in the application.
Why do we need a desktop app, not just a web app?
The web (izziapi.com/aibase) is the place for knowledge sharing and exploring the community graph. However, to actually work with agents — running multiple agents in parallel, maintaining long sessions, manipulating local files, connecting machine tools — a desktop app has advantages: running background processes, accessing the file system, and consolidating multiple "assistants" into one window.
Starizzi is built on a desktop foundation (Electron) to do just that: turn abstract layers (context, skills, agents) into something you can click and run.
The four main blocks of Starizzi
1. Agent Hub — the place to install and run agents
At the center of Starizzi is an Agent Hub: a grid of "Top Agents" for you to install and launch assistants. Notably, there are two types of runtime:
- Local agents: agents running on your machine (usually via Docker) — powerful, private, but require installation.
- Izzi-native agents: agents running through the Izzi API with a persona system prompt — install with one click (instant install), no Docker required. Typical examples are personas like Socrates (the critic/evaluator) and Orchestrator (the coordinator) — which embody the "agent team" thinking discussed in Article 4.
This bifurcation is crucial: it provides newcomers with a frictionless entry (izzi-native, click to run) while still allowing advanced users to run fully-fledged local agents.
2. AI Branching Graph Workspace — thinking in branches
Instead of a linear chat frame, Starizzi features a branching graph workspace similar to Obsidian: each node is an idea/work session, and the AI can branch off the conversation into a graph. You chat per node, using commands like branch / summarize / merge, and the thought tree is preserved.
This embodies the spirit of a "second brain" at the operational level: knowledge and workflows appear as a graph, not as a lost chat line.
3. Knowledge universe — connecting to the community graph
Starizzi embeds the public knowledge graph of izziapi.com (which is /aibase/graph) directly within the app, with the operation to "load" knowledge clusters to start from a map rather than a blank page. This means the personal brain (your working graph) and the community brain (shared knowledge) coexist — reading from the community, working on your own graph.
(The series of articles you are reading, once published, will reside within that very knowledge universe.)
4. Agent Workspace 3 Zones — Operational Cockpit
The workspace is organized into three zones, reflecting the "operational platform" mindset:
- Left — Navigation: session/spaces list.
- Center — Conversation (hero): where you primarily work with the agent.
- Right — Operational Rail: includes a business block (USD/VND balance, package, budget clock, button to purchase additional API), agent list by group, and preset loops (research, automation, coding, content, data/RAG, orchestration) to apply a process with a click.
This "rail" zone directly reflects the governance + cost pillar emphasized in Post 0: you always see how much you are spending and who you are coordinating.
Synchronization with the Backend "Brain"
Starizzi is not an island. Its graph and memory synchronize with the common backend of izziapi.com (the endpoints /api/aibase/*, authenticated by session login). This allows nodes you create on the desktop to appear on the web and vice versa; there is an offline queue mechanism and a "last-write-wins" rule to avoid conflicts when the network is down.
The key point: one brain, multiple surfaces — desktop, web, and (in the future) other channels all read/write to a knowledge graph. This truly embodies the "second brain" principle of the entire platform.
Commercial Platform: Running on Izzi API
Operationally, model calls in Starizzi go through the Izzi API gateway (izziapi.com). You load balances, see the budget clock, and purchase more when needed — a transparent "pay-as-you-go" model. This allows Starizzi to be model-agnostic (routing to multiple providers) while keeping auth + billing in one place.
Starizzi ↔ Agent Development Toolkit
Put together, Starizzi is where six layers come to life:
| Layer (in the chain) | Manifestation in Starizzi |
|---|---|
| CLAUDE.md (memory) | Context per session/node in the graph workspace |
| Skills (expertise) | Loop presets + packaged capabilities of each agent |
| Hooks (control) | Operational guardrails + budget clock |
| Subagents (delegation) | Agent Hub + persona (Socrates, Orchestrator) |
| Plugins (distribution) | One-click agent installation (izzi-native) |
| MCP (connection) | Backend synchronization + external tool integration |
In other words: the chain teaches you the principles; Starizzi gives you a playground to apply them without having to build infrastructure from scratch.
Summary & Next Steps
Starizzi Desktop is the operational cockpit for agents of izziapi.com: an Agent Hub (local agents + izzi-native), a branching graph space for thinking, a window into the community knowledge universe, and an operational rail encompassing both coordination and cost — all synchronized with the common backend "brain."
Having understood the tools, the practical question is: how to use it to operate as a "one-person company" — where you are the sole individual, and the rest is a team of agents? That is Post 9 — the final, most practical post of the series.
References
- izziapi.com — platform & knowledge graph: https://izziapi.com/aibase/graph
- Platform concept (comparison): Anthropic — Steering Claude Code: https://claude.com/blog/steering-claude-code-skills-hooks-rules-subagents-and-more
- (Feature details of Starizzi: see in the latest application.)
Article 8/9 in the "Agent Development Toolkit with Claude" series. Previous ← Article 7: Merging · Next → Article 9: Building a "one-person company" on Starizzi.