ollaAgent

A local LLM agent powered by ollama — with persistent memory, plan mode, and parallel subagents.

Features

Requirements

• Python 3.11+
• ollama running locally (or via Cloudflare Access)
• uv

Installation

git clone https://github.com/github010000/ollaAgent
cd ollaAgent
uv sync

Usage

# Start the agent (default model: qwen3-coder-next:latest)
ollaagent

# With a specific model
ollaagent --model qwen2.5-coder:7b

# With a remote ollama host
ollaagent --host https://your-ollama.example.com

Built-in Commands

Subagent Usage

Single model across all tasks:

/subagent
> --model llama3:8b task one | task two | task three

Per-task model assignment:

> @qwen2.5-coder:7b write a sorting algorithm | @llama3:8b explain it

Configuration

Create a .env file in the project root:

OLLAMA_HOST=http://localhost:11434
CF_ACCESS_CLIENT_ID=
CF_ACCESS_CLIENT_SECRET=

Project Structure

ollaAgent/
├── agent.py          # Main agent loop & CLI entry point
├── memory.py         # Persistent memory (JSON)
├── plan_mode.py      # Plan-only mode (tools=[])
├── subagent.py       # Parallel subagents via multiprocessing
├── tool_bash.py      # Bash tool with permission control
├── permissions.py    # Allow/deny pattern matching
├── config_loader.py  # YAML config & system prompt builder
└── ollama_client.py  # Ollama client factory

Running Tests

uv run pytest

License

MIT — see LICENSE for details.