I built an open-source coding agent that uses a code graph to reduce context, cost, and search time

Описание и применение

Faber – это открытый терминальный агент, использующий граф кода для быстрого поиска нужных файлов, что снижает токен‑потребление и стоимость API‑запросов.

Задача

Developers and students often need AI assistance but face high token costs and slow navigation when the agent scans entire codebases.

Результат

Faber reduces token usage and API costs by up to 50% while speeding up codebase navigation through code‑graph guidance and prompt caching.

Рабочий процесс

• 1. Install Node.js 22.5+ and run npm install -g faberwright. 2. Navigate to a project directory and start Faber with faber. 3. Ask a coding question; Faber queries the code graph to identify relevant files. 4. The agent sends a concise prompt to the chosen LLM (Claude, OpenAI Codex, or a local model). 5. Faber caches the prompt and uses a Chain‑of‑Draft style to generate output, reducing unnecessary tokens. 6. Git integration detects changes, optionally commits per task, and supports /undo and /redo. 7. Use /usage to view expenditure, token usage, and estimated savings.

Доказательства

Faber uses a code graph to find relevant files and symbols first, instead of repeatedly scanning large parts of a repository.

Источник

Prompt caching to reduce repeated input cost

Источник

Git-aware workflow with change detection, optional per-task commits, /undo, and /redo

Источник