Definitions of Maestro-specific terms and concepts.

Core Concepts

TermDefinition
SessionAn independent, checkpointable, resumable workspace containing memories, files, tools, and sandbox. The fundamental unit of work in Maestro.
TurnOne complete cycle of user input, Maestro processing, tool execution, and result presentation. Sessions consist of many turns.
AgentThe AI entity that orchestrates models, tools, and validation to accomplish your goals within a session.
CapacityThe token-based limit on how much context (memories, files, tools) a session can maintain. Visible via capacity indicator in UI.

Memory and Context

TermDefinition
MemoryDialog history and interaction records stored in a session. Includes user requests, agent replies, tool results, and validation outcomes.
Memory CompactionProcess of summarizing old dialog to reduce token usage while preserving key information.
ForgettingSelective removal of turns or memory segments to free capacity and improve focus.
TokenUnit of text that models process. Roughly 3/4 of a word in English. Capacity and costs measured in tokens.
Context WindowThe maximum number of tokens a model can process in one turn. Maestro manages context to stay within limits.

Files and Storage

TermDefinition
FileStoreThe persistent storage system that maintains your session’s files with complete iteration history.
File IterationA snapshot of a file at a particular point in time. Every change creates a new iteration; history is complete.
File ActivationMaking a file’s content visible to Maestro. Activated files consume capacity; others are indexed but not loaded.
FiletreeThe hierarchical structure of all files in your session. Maestro can see the tree without loading all file contents.
Synchronized FilesFiles that sync between session storage and sandbox. Includes text-based source code and certain media types.
CheckpointA point-in-time snapshot of complete session state (agent configuration and FileStore) enabling restoration.

Tools and Capabilities

TermDefinition
ToolA capability Maestro invokes autonomously (e.g., View Files, Execute Command, Complex Reasoning).
CommandA user-invoked action (e.g., /clone, /forget, /tools) that controls session behavior or accesses utilities.
ProposalA suggested code change created by Maestro using PROPOSE_NEW, PROPOSE_EDIT, or PROPOSE_OVERWRITE tags.
Apply ChangesThe tool Maestro must use to commit all pending proposals to the codebase and sandbox.
Complex CodingAdvanced tool with high reasoning budget for implementing sophisticated multi-file code changes.
Complex ReasoningAdvanced tool for deep analytical problems requiring systematic thinking and extended reasoning.

Sandbox

TermDefinition
SandboxAn isolated Ubuntu Linux execution environment where code runs, tests execute, and validation occurs.
Terminal WindowA named shell session within the sandbox. Multiple windows allow parallel processes and organized execution.
Terminal StreamerUI component showing real-time output from sandbox terminal commands.
Preview URLPublic URL exposing sandbox port 8080, allowing access to web services running in sandbox.
Sandbox InstanceA specific sandbox configuration. Sessions can have multiple instances (default, custom resources, privileged, SSH).

Source Control

TermDefinition
CloneDownloading a repository from GitHub into your session, making files immediately available to Maestro.
Overwrite CloneRe-cloning an already-cloned repository to pull remote changes, creating new iterations for changed files.
Pull Request (PR)A GitHub feature requesting merge of changes from feature branch to base branch. Maestro can create and update PRs.
Feature BranchA git branch containing changes for a specific feature, typically merged via PR.
Base BranchThe target branch for a PR, usually main or develop.
Auto-CloneAutomatic cloning of newly created repository immediately after creation.

Validation and Quality

TermDefinition
ValidationProcess of proving code correctness through tests, benchmarks, and analysis. Evidence-based, not assumption-based.
BaselineMeasurement of current system performance before optimizations, used for comparison after changes.
BenchmarkSystematic measurement of performance (latency, throughput, resource usage) under defined conditions.
Test CoveragePercentage of code exercised by tests. Higher coverage indicates more thorough testing.
RegressionNew bug introduced by changes, or performance degradation. Quality gates prevent regressions.
Edge CaseUnusual or extreme scenario that code must handle correctly. Comprehensive testing covers edge cases.
Quality GateValidation checkpoint that must be passed before proceeding (e.g., all tests must pass before creating PR).
Smoke TestQuick, basic test verifying fundamental functionality works after deployment or major change.
Regression TestTest verifying that previously working functionality still works after changes.
Integration TestTest verifying correct interaction between components or services.
End-to-End Test (E2E)Test validating complete user workflow from start to finish.

Credentials and Security

TermDefinition
CredentialSecret information (API key, token, password) needed to access external services.
Credential ActivationMaking a credential available as environment variable in sandbox for a specific session.
Credential ManagerUI for registering, editing, and managing credentials. Accessed via top-right menu.
OAuthAuthentication protocol allowing secure authorization without sharing passwords. Used for GitHub, GitLab, etc.
Session-ScopedLimited to one session only. Credentials are session-scoped (not shared across sessions).

Billing

TermDefinition
CreditsConsumable units purchased to pay for model usage, sandbox time, and storage. Different operations cost different credit amounts.
Auto-ReloadAutomatic credit purchase when balance drops below threshold, preventing session interruption.
Plan TierSubscription level (Basic, Pro, Ultimate) determining features, resource limits, and included credit allocation.
Usage AnalyticsDetailed breakdown of credit consumption by category (models, sandbox, storage).

Workflow Patterns

TermDefinition
Goal-Driven RequestStructuring requests around desired outcomes rather than implementation steps.
Evidence-Based ValidationRequiring proof (test output, benchmarks, logs) for all claims rather than accepting assertions.
Test-First Development (TDD)Writing comprehensive tests before implementation, then implementing to satisfy tests.
Benchmark-Driven OptimizationEstablishing baseline performance, implementing changes, measuring improvements with same test harness.
Specification-Driven DevelopmentCreating complete technical specification before implementation, then implementing according to spec.

Common Abbreviations

AbbreviationFull Name
APIApplication Programming Interface
CI/CDContinuous Integration/Continuous Deployment
CPUCentral Processing Unit
CRUDCreate, Read, Update, Delete
CSSCascading Style Sheets
E2EEnd-to-End (testing)
GPUGraphics Processing Unit
HTMLHyperText Markup Language
HTTP/HTTPSHyperText Transfer Protocol (Secure)
IDEIntegrated Development Environment
I/OInput/Output
JSONJavaScript Object Notation
JWTJSON Web Token
LLMLarge Language Model
OAuthOpen Authorization
ORMObject-Relational Mapping
PRPull Request
RAMRandom Access Memory
RESTRepresentational State Transfer
SDKSoftware Development Kit
SQLStructured Query Language
SSHSecure Shell
TDDTest-Driven Development
TLSTransport Layer Security
TUIText User Interface
UI/UXUser Interface/User Experience
URLUniform Resource Locator
VMVirtual Machine
VPNVirtual Private Network
WIPWork In Progress
YAMLYAML Ain’t Markup Language