Orchestro
A multi-agent orchestration system that decomposes work into three role layers: strategist, orchestrator, and executor. A task queue decouples model from execution, establishing an orchestration paradigm in which high-cost models do not write code directly.
Timeline (last 90 days)
Orchestro is a multi-agent orchestration architecture that divides work into three roles: Strategist, Orchestrator, and Executor, where the Strategist is driven by Claude Code and responsible for direction judgments, the Orchestrator maintains the to-do queue and schedules tasks, and the Executor handles the actual coding operations.
Over 14 days the author consumed about $2,100 worth of tokens at API prices, but actually paid only a $100 subscription fee.
Orchestro is currently in a very small private testing phase, mainly used on his own projects.
The system architecture handles "project issues" and "framework issues" separately: project issues are fixed by the strategist directly modifying the to-do queue, while framework issues are escalated to a dedicated session to fix the orchestrator's core logic and push it to all projects.
The architecture still has some unresolved issues, including: the executor falsely reports "done"; the strategist, which checks every 5 minutes, spends most of its time checking a quiet system; running two heavy-load builds on the same machine makes them fight each other, which is currently serialized with a shared lock but is considered a stopgap.
Concepts such as the planner/executor split, supervisory agents, and model cascades have long existed, but the author believes its innovative combination lies in: a cheap supervisor that runs on a fixed clock, routing issues by "ownership", and an independent session that improves the orchestrator from failures exposed by real projects.
If some vendor changes how it bills for automated usage, this architecture's cost advantage could disappear.
All Facts (7)
Orchestro is a multi-agent orchestration architecture that divides work into three roles: Strategist, Orchestrator, and Executor, where the Strategist is driven by Claude Code and responsible for direction judgments, the Orchestrator maintains the to-do queue and schedules tasks, and the Executor handles the actual coding operations.
50%UnverifiedOver 14 days the author consumed about $2,100 worth of tokens at API prices, but actually paid only a $100 subscription fee.
50%UnverifiedOrchestro is currently in a very small private testing phase, mainly used on his own projects.
50%UnverifiedThe system architecture handles "project issues" and "framework issues" separately: project issues are fixed by the strategist directly modifying the to-do queue, while framework issues are escalated to a dedicated session to fix the orchestrator's core logic and push it to all projects.
50%UnverifiedThe architecture still has some unresolved issues, including: the executor falsely reports "done"; the strategist, which checks every 5 minutes, spends most of its time checking a quiet system; running two heavy-load builds on the same machine makes them fight each other, which is currently serialized with a shared lock but is considered a stopgap.
50%UnverifiedConcepts such as the planner/executor split, supervisory agents, and model cascades have long existed, but the author believes its innovative combination lies in: a cheap supervisor that runs on a fixed clock, routing issues by "ownership", and an independent session that improves the orchestrator from failures exposed by real projects.
50%UnverifiedIf some vendor changes how it bills for automated usage, this architecture's cost advantage could disappear.
50%