Chiggabot
Local AI AgentsFully local AI agent with RAG, code execution, and multi-language runtime.
The Problem
Integrating cloud-based LLM APIs compromises codebase privacy, creates recursive API subscription bills, and increases single-point failure dependencies.
System Architecture
User Query -> Embedding Model -> Vector Database -> Local LLM -> Response Generation & Execution Sandbox.
+-------------------+ +----------------------+ +----------------------+
| User Query Input | ---> | sentence-transformer | ---> | FAISS / Chroma Vector|
| (Local Console) | | Local Embeddings | | Database Triage |
+-------------------+ +----------------------+ +----------------------+
|
v
+-------------------+ +----------------------+ +----------------------+
| Isolated Docker | <--- | Response Synthesizer | <--- | Ollama Offline LLM |
| Execution Sandbox | | & Text-to-Speech | | (Zero cloud boundary)|
+-------------------+ +----------------------+ +----------------------+Core Implementation
Built an entirely local agent engine powered by Ollama local models. Configured a custom Retrieval-Augmented Generation (RAG) pipeline leveraging FAISS and Chroma vector stores for codebase scanning. Added local text-to-speech voice synthesis and a secure execution sandbox using Docker containers.
Security Considerations
Script evaluation occurs in an isolated Docker runtime environment with capped resources and blocked networking, preventing malicious code from accessing the host OS filesystem or local network.
Quantifiable Impact
Created a private, zero-subscription development assistant completely independent of internet access or third-party servers.
Project Links
Internal corporate deployment.