Jose Gibson
← All Projects

Jerry — AI Agent Framework

CLI tool that turns any local directory into an AI-powered workspace by indexing its documents into a vector store and exposing a LangGraph-backed, tool-using agent.

PythonLangGraphRAG with vector store (`.md` and `.txt` ingestion)Extensible ToolProvider systemState stored in hidden `.jerry/` folder
Keeping vector store state fresh when project files change wScoping agent memory to the directory while supporting neste

Portfolio Highlights

  • Built a Python CLI that converts any project directory into a context-aware AI workspace using LangGraph and document RAG.
  • Designed a modular ToolProvider system enabling extensible tool registration with isolated per-directory state.

Snapshot

  • Period: October 2025
  • Source: `github.com/josegibson/Jerry`
  • Domain: AI agents, RAG, developer tooling
  • Status: Public release

Stack

  • Python
  • LangGraph
  • RAG with vector store (`.md` and `.txt` ingestion)
  • Extensible ToolProvider system
  • State stored in hidden `.jerry/` folder

What I Built

  • Directory-scoped agent instantiation: run `jerry` in any folder to spin up a context-aware agent.
  • Auto-indexing pipeline that vectorizes readable documents at startup.
  • Modular ToolProvider architecture allowing new tools to be registered without touching core agent logic.
  • State isolation so each workspace keeps its own `.jerry/` data — no cross-project bleed.
  • Full documentation suite: INSTALL, QUICKSTART, USER_GUIDE, CONTRIBUTING.

Development Timeline

  1. Implemented initial directory-scoped agent with basic RAG indexing.
  2. GoalAdd extensible ToolProvider architecture
  3. FixedDesigning a ToolProvider interface that is extensible without requiring boilerplate for simple tools.
  4. Added public entry
  5. Another public entry