Architecture

High-level design of Stella: how the agent loop, tools, and supporting services fit together. Start here to understand the system before diving into code.

🚧 Starter page — replace the placeholders below with real content and diagrams.

System overview

Add a system diagram (Mermaid works in GitHub wikis) and a one-paragraph narrative of the request lifecycle: user prompt → agent loop → tool execution → response.

flowchart LR
    U[User / IDE] --> O[Orchestrator]
    O --> L[LLM Provider]
    O --> T[Tool Runtime]
    T --> S[(Sandbox)]
    O --> M[(State / Memory)]

Loading

Components

Component	Responsibility	Code
Orchestrator	Runs the agent loop, manages turns and interrupts	link to package
Tool runtime	Executes file/shell/search tools in the sandbox	link to package
Context manager	Builds prompts, handles truncation and memory	link to package
Eval harness	Regression suites for agent behavior	link to package

Key design decisions

Link the ADRs that shaped the current architecture — see Decision Records.

Related pages

• Agent Design — prompt and tool-level detail
• Operations & Runbooks — how this runs in production