Skip to content

Ollama Agent

License Python 3.11+ DeepAgents LangChain Ollama

Ollama Agent is an autonomous local AI assistant that combines an interactive terminal REPL and a non-interactive CLI with deep system tools, subagent delegation, vector-based RAG, long-term memory, and full Model Context Protocol (MCP) support. Built on top of DeepAgents and LangChain, it runs entirely on your local machine using Ollama models while guaranteeing privacy, local execution control, and real-time streaming feedback.

Quick Installation

Install Ollama Agent instantly in an isolated Python environment using pipx: 

pipx install git+https://github.com/arrase/ollama-agent.git

Key Features

Interactive REPL

Modern terminal UI with live Markdown rendering, history navigation, interactive slash commands, mid-session model switching, and prompt autocompletion.

Shell Execution

Integrated local shell backend with configurable directory traversal, sandboxing, environment inheritance, and customizable execution timeouts.

Deep RAG Engine

Local vector search database powered by Ollama embeddings and Qdrant backend for semantic document retrieval and context-aware answers.

Persistent Memory

Native memory layer backed by MEMORY.md, enabling long-term cross-session knowledge retention and persistent project context.

Custom Subagents

Define specialized subagents in settings.yaml with isolated context windows, specialized system prompts, dedicated models, and MCP tool sets.

MCP Integration

Native support for Model Context Protocol servers to dynamically extend agent capabilities with external tools, APIs, and databases.

Agent Skills

Reusable, modular skill definitions adhering to the Agent Skills spec with progressive disclosure and dynamic context loading.

Task Management

Save, configure, and re-execute multi-step prompts with custom model choices and reasoning effort parameters on demand.

HITL & YOLO Mode

Human-in-the-Loop approval workflow for destructive shell & file operations with optional zero-friction YOLO mode toggle.

Interactive REPL Non-Interactive Prompt
Interactive REPL UI Non-Interactive CLI
Full REPL interface with markdown rendering, status bar & tool calls Single prompt execution from CLI with structured output

Documentation Index

Explore the complete technical guides for Ollama Agent:

  • System Architecture: Graph orchestration, SQLite state persistence, streaming parsers, shell middleware, and reasoning trace capture.
  • CLI & REPL User Guide: Terminal interface commands, keyboard shortcuts, @-file mentions, multiline inputs, and non-interactive usage.
  • MCP & Subagents Architecture: Model Context Protocol setup, subagent configuration, and isolated task delegation.
  • RAG Engine Guide: Local vector database creation, document chunking, embeddings setup with Ollama & Qdrant.
  • Skills, Tasks & Memory: Authoring reusable skills, managing saved task templates, and configuring persistent MEMORY.md.
  • Configuration & Tracing: Comprehensive settings.yaml reference, context window auto-resolution, and model reasoning effort levels.