ollama-agentic

A beautiful, agentic terminal interface for Ollama — run local LLMs with auto tool-calling, long-term memory, git integration, concurrent subagents, and semantic code search.

⚠️ Requirement: Ollama must be installed first

This CLI is a frontend for Ollama. It will not work without Ollama installed and running on your machine.

1. Download and install Ollama from ollama.com/download
2. Start it: ollama serve (or open the Ollama desktop app)
3. Pull a model: ollama pull mistral or ollama pull llama3.1:8b

Then install and launch this CLI:

pip install ollama-agentic
ollama-cli

Features

• ⚡ Auto mode — model autonomously calls tools to complete tasks (/auto)
• 🐝 Swarm agents — /swarm splits complex tasks across parallel background agents
• 🔍 Semantic code search (RAG) — AST-aware local codebase indexing, no API needed
• 🌿 Git integration — /git status, diff, log, commit (with AI messages), branch, stash
• 🔁 Iterative debug loop — /run file.py auto-fixes errors until code passes
• 📋 Plan executor — /plan <goal> breaks goals into typed steps and executes them
• 🧠 Long-term memory — /remember stores facts that persist across sessions
• ⬇️ Arrow-key model picker — /install lets you browse and download 25+ models
• 🔧 Agent tools — /shell, /file, /fetch, /ls inject real context into chats
• 💾 Conversation saving — /save and /load persist chats as JSON
• 🎭 Personas — save and load system prompt presets
• 🆚 Compare mode — run the same prompt through two models side by side

Usage

ollama-cli                       # start chatting
ollama-cli --model qwen2.5:7b    # start with a specific model
ollama-cli --auto                # start in autonomous agent mode
ollama-cli --compare             # compare two models side by side

Commands

Chat & Navigation

Models

Agentic

Git

RAG — Semantic Code Search

Memory

Context Injection

Conversations & Personas

Swarm Agents

/swarm decomposes a complex task into independent subtasks and runs them as parallel agents in the background. You keep using the CLI while they work.

you › /swarm research React Server Components vs traditional SSR
you › /swarm-status          # check mid-task
you › /swarm-status full     # read each agent's full output

RAG — Semantic Code Search

Run from inside any git repo. Uses AST-aware chunking for Python and sliding-window chunking for all other languages. Embeddings run fully offline via sentence-transformers.

RAG dependencies are optional — the CLI works fine without them:

pip install lancedb sentence-transformers tree-sitter tree-sitter-python

you › /rag index             # index your project (~seconds)
you › /rag search auth flow  # semantic search
you › /rag auto              # auto-inject relevant chunks into every chat

The index lives in .ollama_rag/ inside your project. Only changed files are re-indexed on subsequent runs.

Agent Mode

Toggle with /auto or launch with --auto. The model calls tools, reads results, and loops until the task is done.

⚡ you › look at main.py and find any bugs
⚡ you › write a web scraper for hacker news and run it
⚡ you › set up a basic Flask app in this folder

Config & Data

Requirements

• Python 3.10+
• macOS, Linux, or Windows
• Ollama installed and running — ollama.com/download

Roadmap

• Project memory — /understand deep-reads your codebase and stores structured knowledge
• MCP server support — connect to filesystem, GitHub, Postgres, browser tools
• TUI dashboard — split-pane interface with live swarm agent view
• API key integrations — Claude, OpenAI, Gemini, Groq as model backends

Contributing

PRs and issues welcome at github.com/Akhil123454321/ollama-cli. Keep changes focused and include tests where appropriate.

License

MIT — see LICENSE