Ollama Agent - Local AI Assistant

Key Features

Interactive REPL

A modern terminal-based chat interface with rich Markdown support.

Slash Commands

/new	Start a fresh session context
/history	View list of past sessions
/model-set <name>	Switch active Ollama model
/clear	Clear the current screen
/exit	Close the application

Non-Interactive Mode

Execute single-shot tasks directly from your shell.

CLI Args

-p "..."	Prompt to execute immediately
--rag <db>	Use specific RAG database
--skills-dir <dir>	Add extra skills directory

# Example ollama-agent -p "Summarize README.md" --rag docs_db

Deep RAG

Contextual Retrieval Augmented Generation using local vector stores.

Management Commands

/rag-create <name>	Initialize a new knowledge base
/rag-add <path>	Ingest file or directory
/rag-load <name>	Activate a specific DB

DeepAgents Memory

Long-term persistence powered by DeepAgents' native memory layer.

The agent automatically stores and retrieves user preferences and context across sessions.

Storage

MEMORY.md

Markdown file where facts are persisted

Screen Vision

Give your agent eyes. Capture and analyze your screen contents.

Supported on Linux (X11/Wayland).

Syntax

@dp0	Capture primary display
@dp1	Capture second monitor

# Example prompt "Look at @dp0 and help me fix this error message"

Task Automation

Define reusable workflows in YAML to automate complex queries.

Task Commands

/tasks	List available tasks
/task-run <id>	Execute a specific task
/task-create	Interactive task builder

# ~/.ollama-agent/tasks/analyze.yaml title: Code Analysis model: codellama prompt: | Analyze the current file for security vulnerabilities and suggest improvements.

MCP Support

Extend capabilities with Model Context Protocol servers.

Each configured MCP server is exposed as a delegation tool (use_<name>) backed by its own DeepAgents sub-agent using langchain-mcp-adapters. Edit ~/.ollama-agent/mcp_servers.json to add servers.

{ "mcpServers": { "filesystem": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-filesystem", "/home/user"] } } }

Skills

Reusable, on-demand agent capabilities following the Agent Skills specification.

Skills provide task-specific instructions via progressive disclosure — the agent only reads full instructions when a skill’s description matches the prompt, keeping the system prompt lean.

Skill Commands

/skills	List all available skills
/skill-show <id>	View full skill details
/skill-create <id>	Interactive skill builder
/skill-delete <id>	Remove a skill

# ~/.ollama-agent/skills/langgraph-docs/SKILL.md --- name: langgraph-docs description: Fetch relevant LangGraph documentation. --- # langgraph-docs ## Instructions 1. Fetch the documentation index. 2. Select relevant pages. 3. Provide accurate guidance.

Integrated Shell Execution

The agent interacts with your system through a built-in shell middleware.

Shell commands are handled transparently via ShellToolMiddleware, removing the need for an explicit execute_command tool. The middleware integrates directly with the DeepAgents graph for seamless tool execution.

Built-in Tools

Shell Middleware	Transparent command execution with configurable timeout
rag_search	Search loaded RAG knowledge bases

# Example prompt "Find all Python files in src/ modified in the last 24h"

Getting Started

Requirements

Ensure you have Ollama running with a tool-capable model (like llama3.1) and an embedding model.

ollama pull gpt-oss:20b
ollama pull nomic-embed-text

Installation

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

Interactive Mode

Start the chat interface to begin a session.

ollama-agent

One-off Commands

Execute a single prompt directly from your shell.

ollama-agent -p "Find large files in /var/log and summarize them"

Your Local AI Assistant