chippr-agi
A TypeScript agent framework, rewritten around an entity-component-system architecture, that gives autonomous agent workflows persistent SQLite-backed memory with semantic search, a choice of cloud or local reasoning models, and containerized execution.
chippr-agi v2 is a ground-up rewrite of an earlier BabyAGI-style proof of concept, built on the premise that agent orchestration fits an entity-component-system model: tasks are entities, data lives in components, and systems subscribe to a typed event bus to act on them. Entities, components and events persist in SQLite, and systems are loaded dynamically and self-register from a directory of source files.
It supports Claude as a primary reasoning engine with full tool use, or local models through Ollama, llama.cpp or BitNet over an OpenAI-compatible API for offline or cost-sensitive work. A semantic memory layer embeds stored memories with Google's Gemini embedding models and retrieves them by cosine similarity, using task-typed embeddings for documents versus queries. Built-in systems route tasks to the right handler, break objectives into subtasks, score task completion, and run tasks on cron schedules, and each agent context runs isolated in its own Docker or Apple container.