About Interactive Prompts
Maestro can pause execution and request your input through interactive prompts. This is not a limitation — it is a design choice that keeps you in control of decisions that should not be automated.
Interactive prompts come in several forms: text input for clarification, file upload requests, multiple-choice buttons for selecting approaches, and interactive tables for complex selections like credential management or file selection.
Prompts persist across browser refreshes and synchronize across devices. If Maestro asks a question while you are on your desktop, you can answer it from your phone. This reflects the reality that knowledge work does not happen at a single desk.
About Memory Management
Session memory is not unlimited, and this is by design. An unbounded conversation history would degrade quality as the model struggles to process increasingly irrelevant context.
Compaction summarizes old dialog, preserving key decisions while reducing token usage. It is analogous to taking notes during a meeting rather than recording the entire conversation.
Forgetting removes dialog selectively. Failed approaches, debugging iterations, and exploratory dead ends can be removed once their lessons are captured.
Refreshing resets file view to show only the latest iteration. This is particularly impactful when you have been iterating heavily on files, accumulating dozens of iterations that consume capacity.
The philosophy is intentional curation, not hoarding. A session with a clean, focused context produces better results than one burdened with everything that ever happened.
About Multi-Device Support
Sessions synchronize across all your devices. This reflects how developers actually work: starting at a desktop, checking progress on a phone during a break, continuing on a laptop at a coffee shop.
What synchronizes: complete dialog history, file state, tool execution results, interactive prompts, and session settings.
Collaborative sessions — multiple people viewing and interacting with the same session — are possible with appropriate access controls. This enables patterns like a team lead defining requirements while a developer provides domain expertise, both seeing identical state.
About Chaos Testing
Chaos testing injects random failures into the sandbox — network issues, crashes, resource exhaustion — to test system resilience. This is an experimental feature that validates error handling and recovery patterns.
The concept comes from production reliability engineering. Systems that handle failure gracefully in controlled testing are more likely to handle failure gracefully in production. Chaos testing turns this principle into an automated validation tool.
About Advanced Sandbox Configurations
The default sandbox (2 vCPU, 7 GB RAM, Ubuntu Linux) handles most workloads. Advanced configurations serve specialized needs:
High-memory sandboxes (up to 64 GB) for data processing and memory-intensive computations. GPU sandboxes for machine learning workloads. Privileged sandboxes that support Docker for containerized workflows. SSH sandboxes that connect to external systems for deployment operations or remote debugging.
These configurations exist because real engineering work spans a wide range of resource requirements. The sandbox adapts to the work, not the other way around.
About Skills
Skills are reusable packages that give Maestro expertise in specific domains. They represent externalized knowledge — frameworks expertise, quality standards, specialized automation — that would otherwise need to be explained in every session.
Skills work in the background, activating automatically when they detect relevant work. This means Maestro “just knows more” about the domains covered by your enabled skills, without requiring explicit invocation.
The skill ecosystem is open: public repositories provide community-contributed capabilities, organizations create internal skills for company-specific standards, and individuals can create skills for their specific needs.
Further Reading
