🦙 OllamaCode

A Ollama Code CLI powered by Ollama local models — fully offline, private, and free.

Run an AI coding agent in your terminal that reads files, writes code, executes commands, and iterates autonomously — all using local models via Ollama.

Features

• Agentic loop — AI plans and executes multi-step tasks autonomously
• Full tool suite — read/write/edit files, run bash, grep, find, git, and more
• Streaming output — see responses as they're generated
• Tool approval — review every tool call before it runs (or enable auto-approve)
• Model picker — switch between any installed Ollama model at runtime
• Conversation history — full context across turns
• Slash commands — /help, /model, /clear, /cd, /add, and more
• Non-interactive mode — pipe prompts in for scripting
• Markdown rendering — formatted output in terminal
• Line history — up/down arrows navigate past inputs

Installation

Prerequisites

1. Node.js 18+ — https://nodejs.org
2. Ollama — https://ollama.ai

# Install Ollama (macOS/Linux)
curl -fsSL https://ollama.ai/install.sh | sh

# Or on macOS with Homebrew
brew install ollama

Install OllamaCode

# Clone or download the project
cd ollamacode

# Install dependencies
npm install

# Build
npm run build

# Install globally
npm install -g .

# Or run directly
npm run dev

Pull a recommended model

OllamaCode works best with code-focused models:

# Best overall for coding (recommended)
ollama pull qwen2.5-coder:7b

# Larger, more capable
ollama pull qwen2.5-coder:14b

# DeepSeek (excellent at code)
ollama pull deepseek-coder-v2

# Meta's code model
ollama pull codellama:13b

# Fast and lightweight
ollama pull llama3.2:3b

Usage

Start interactive session

ollamacode
# or short alias:
oc

Use a specific model

ollamacode --model qwen2.5-coder:14b

Change working directory

ollamacode --dir /path/to/your/project

Non-interactive / pipe mode

# Single prompt
ollamacode --print "explain the main function in src/index.ts"

# Pipe from stdin (not yet, use --print)
oc -p "write unit tests for utils.ts"

Auto-approve all tools (for scripting)

ollamacode --auto-approve -p "refactor all console.log to use a logger"

Slash Commands

| Command | Description | |---|---| | /help | Show all commands | | /exit or /quit | Exit OllamaCode | | /clear | Clear conversation history | | /model [name] | Show or switch model | | /models | List installed Ollama models | | /pull <name> | Pull a new Ollama model | | /config | Show current config | | /set <key> <value> | Update a config value | | /approve on\|off | Toggle auto-approval of tool calls | | /verbose on\|off | Toggle verbose tool output | | /cd <path> | Change working directory | | /pwd | Show current directory | | /add <file> | Add a file to conversation context | | /history | Show last 10 conversation turns | | /undo | Show last file change |

Config keys for /set

/set model qwen2.5-coder:7b
/set temperature 0.1
/set maxTokens 4096
/set autoApprove true
/set verbose true
/set ollamaUrl http://localhost:11434

Tools Available

The AI agent has access to these tools:

| Tool | Description | |---|---| | read_file | Read a file with line numbers (supports line ranges) | | read_multiple_files | Read several files at once | | write_file | Create or overwrite a file | | edit_file | Surgical text replacement in a file | | list_directory | Tree view of directory | | bash | Run shell commands | | grep | Search files with regex | | find_files | Find files by name pattern | | delete_file | Delete a file | | move_file | Move or rename a file | | git_status | Show git status and recent commits |

Configuration

Config is stored at ~/.ollamacode/config.json.

{
  "ollamaUrl": "http://localhost:11434",
  "model": "qwen2.5-coder:7b",
  "temperature": 0.2,
  "maxTokens": 8192,
  "autoApprove": false,
  "verbose": false,
  "maxFileSize": 1048576
}

Example Tasks

❯ Create a REST API in Express with CRUD endpoints for a todos resource

❯ Read package.json and set up ESLint + Prettier with my existing config

❯ Find all TODO comments in the codebase and create a TODOS.md file

❯ Write unit tests for every function in src/utils.ts using Vitest

❯ Refactor this class to use the repository pattern

❯ Fix the TypeScript errors in src/

❯ Add input validation to all API endpoints using Zod

❯ Explain what this codebase does, then suggest improvements

Model Recommendations

| Model | Size | Best For | |---|---|---| | qwen2.5-coder:7b | ~4.7GB | Best balance of speed + quality | | qwen2.5-coder:14b | ~9GB | Higher quality, slower | | deepseek-coder-v2 | ~9GB | Excellent reasoning | | codellama:13b | ~7.4GB | Good all-rounder | | llama3.1:8b | ~4.7GB | Good general purpose | | llama3.2:3b | ~2GB | Fastest, less capable | | phi3:mini | ~2.3GB | Very fast, decent quality |

Note: Tool calling (function calling) support varies by model. qwen2.5-coder has the best tool call support. If a model doesn't support tool calls natively, OllamaCode will still work but may have reduced agentic capability.

Architecture

ollamacode/
├── src/
│   ├── index.ts          # CLI entry point + REPL loop
│   ├── types.ts          # Shared TypeScript types
│   ├── config.ts         # Config load/save
│   ├── ollama.ts         # Ollama API client (streaming)
│   ├── core/
│   │   ├── agent.ts      # Agentic loop + tool execution
│   │   └── commands.ts   # Slash command handlers
│   ├── tools/
│   │   └── index.ts      # All tool implementations
│   └── ui/
│       ├── renderer.ts   # Terminal colours + markdown
│       └── input.ts      # Line editor + history
├── package.json
└── tsconfig.json

Troubleshooting

"Ollama is not running"

ollama serve

"Model not found"

ollama pull qwen2.5-coder:7b

Tool calls not working Some models don't support the Ollama tool call format. Switch to a model that does:

/model qwen2.5-coder:7b

Slow responses Try a smaller model:

/model llama3.2:3b

Or reduce context:

/set maxTokens 2048

Remote Ollama server

/set ollamaUrl http://192.168.1.100:11434

License

MIT