Fleet state
Seven agents report into a shared view with real-time event streaming.
Case study
Internal Operating System
Ongoing← All case studies
Workflow receipt · prettyfly os
PrettyFly OS gives internal agent work one operating surface.
PrettyFly needed a way to see agent state, cost, failure, and handoff without chasing logs across servers.
PrettyFly.aiPrettyFly OScoded proof surface · future motion slot
No. 02
Inspectable proof
Proof videoprettyfly.ai / prettyfly-osPrettyFly OS
No. 03
Original drag
Drag named firstOriginal drag
The work was distributed, but the operating state was not.
Agents were running across servers, but health, cost, tasks, and handoffs were split across logs and local state. The system needed one operating surface before more agent work could be trusted.
- 01
Agent health lived outside a shared operating view.
- 02
Failures could hide until someone checked the wrong place.
- 03
Cost controls were separate from the work being done.
- 04
Task handoffs had no common surface.
No. 04
Operating route
Mapped pathOperating Route
A PrettyFly OS route for agent work: observe, budget, trip, trace, hand off.
PrettyFly built an internal operating surface with fleet state, real-time events, circuit breakers, cost visibility, tracing, and MCP tools for development sessions.
Circuit breakers
Repeated faults trip the route and make recovery explicit.
Cost readout
Per-agent spend, daily budgets, tracing, and monthly reports sit beside the work state.
18 MCP Tools
Observability and task tools are exposed directly to development sessions.
Small runtime
A raw Node.js HTTP server keeps the coordination layer inspectable.
Two auth modes
Supabase JWT handles dashboard access; bearer tokens handle machine communication.
No. 05
Shipped system
ReceiptShipped System
A live PrettyFly OS surface for internal agent operations.
7Agents Running 24/7
$50Monthly Cost
18MCP Tools
0NPM Dependencies
Stack
Node.jsPostgreSQLLangfuseSSEsystemdSupabase Auth
Your route
Bring the workflow before the build.
Name one route that feels expensive. The diagnostic reads the drag, owner, risk, and next move before anyone recommends a build.