🛠 Qarinah: Radically reducing context costs for AI agents
Qarinah has been introduced — a local tool for memory management and context compilation for agents such as Codex and Claude Code. Thanks to the cited packs mechanism, the project reduces the volume of transmitted context by 98.71%, forming a compressed dataset with references to the original sources.
🌍 It solves the problem of context window bloat and high token costs. It allows agents to work effectively with massive codebases, maintaining accuracy and reducing API costs without the need for external vector databases.
👤 It allows developers to save significantly on subscriptions and API usage by passing only the necessary information to the agent, with the ability to verify every fact through direct links to the code.
Source 1: https://github.com/AjnasNB/qarinah