deltron 0.1.1

del the funky robosapien — a terminal AI agent

del — the funky robosapien

A terminal AI agent powered by Ollama and agno. Start an interactive REPL with tools for shell, file operations, math, and web search.

Installation

pip install deltron

Usage

del

This starts an interactive REPL where you can chat with the agent. Type exit, quit, or bye to quit.

Features

• Local Ollama backend — runs models on your machine (no cloud API)
• Shell integration — agent can run commands, manipulate files
• Session persistence — conversation history tied to your shell session
• Knowledge base — automatically load markdown files from ~/.local/share/del/knowledge/
• Symlink support — knowledge discovery follows symlinks and recurses directories
• Project context — injects pwd, git branch, and optional AGENTS.md instructions into the agent

Configuration

Knowledge Base

Create markdown files in ~/.local/share/del/knowledge/ and the agent will automatically load them on startup. Example:

mkdir -p ~/.local/share/del/knowledge
cat > ~/.local/share/del/knowledge/my-notes.md <<EOF
# My Project

- Task 1
- Task 2
EOF

The knowledge base supports symlinks and recursive directory traversal.

Project Instructions (AGENTS.md)

Create an AGENTS.md file in your repo root. The agent will find it and inject its contents into the system prompt:

cat > AGENTS.md <<EOF
# Development Guidelines

- Use test-driven development
- Prefer small commits
- Run tests before pushing
EOF

Architecture

• deltron — the Python module (entry point: deltron.cli:main)
• roles — agent role definitions in Markdown with YAML front matter
• tools_registry — maps tool names to agno toolkits
• context — injects runtime context (cwd, date, git branch, AGENTS.md)
• knowledge — discovers and loads markdown files for RAG
• agent — builds and configures the agent
• cli — CLI entry point

Extending with Custom Roles

Create a new role file in src/deltron/roles/:

---
tools:
  - shell
  - file
  - calculator
model: qwen3.5:latest
---
You are a financial advisor. Help users with investment questions.

Then load it in code with load_role("financial-advisor").

Requirements

• Python 3.11+
• Ollama with at least one model running locally
• 4GB+ RAM