Executor

Crew Memory Graph contains three agents: Planner, Researcher, and Executor.

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.

SkillScaffold contains two roles, Generator and Executor, which are respectively responsible for producing skills and executing tasks, and the two are jointly trained.

Skill internalization refers to the Executor gradually absorbing the behavioral patterns carried by external textual skills into its own policy during training, so that at inference time it can complete the same type of task even without skills being provided.

Executor can automatically infer default policies from interface specifications; for example, GET requests are set to always allow by default.

Executor offers four running modes: local CLI, desktop client, hosted on Executor Cloud, and self-hosted Docker deployment or hosted on Cloudflare Worker.

Installing Executor locally requires Node.js version 20 or above, and can be completed using package managers such as npm or bun.

Executor supports connecting to Agents via HTTP or stdio; the latter only requires running a single NPX command in the terminal.

The Executor usage process is divided into four steps: add integration, create connection, set policies, and connect Agent.

After installing Executor, running executor install registers it as a persistent background service, and running executor web opens the web management page.

Executor is a unified proxy layer sitting between multiple Agents and the outside world, essentially a centralized MCP management and configuration tool.

In addition to MCP Server, Executor also supports OpenAPI, GraphQL, and Google Discovery—as long as an interface can be described using JSON Schema, it can be uniformly converted into a tool callable by an Agent.

Executor provides three permission policy tiers for each tool: always allow, execute after approval, and block directly.

The core mechanism by which FLARE achieves underlying decoupling is a communication abstraction layer and an Executor plugin system, where business logic is encapsulated in platform-agnostic Python components, and process startup, network communication, and resource requests are handled by the corresponding platform adapters.