offcode 0.1.0

Offline AI coding assistant powered by Ollama
offcode-0.1.0 is not a library.

offcode

      ╭──────────╮
      │  ◉    ◉  │   offcode  model:gemma4:e4b
      │   ╰──╯   │   offline coding assistant
      ╰──────────╯   powered by ollama · type a prompt to begin

Offcode ASCII art

What is this?

This is a commandline based coding agent that it's really 100% offline, it just connect to ollama locally. I created this this because projects like opencode makes tcp connections even using offline mode. Use your favourite Ollama model for coding your private projects. No cloud. No API keys. No telemetry. Just you and your model.

Works like Claude Code or OpenCode but talks only to a local Ollama instance but keeping your privacy.

Features

  • Full TUI — scrollable chat interface with live streaming output
  • Tool use — reads/writes files, runs shell commands, searches code
  • Agentic loop — chains multiple tool calls autonomously until the task is done
  • Think support — shows or hides Qwen3/Deepseek reasoning tokens
  • Config file — persistent settings at ~/.config/offcode/config.toml
  • Pure Rust — minimal dependencies, static Linux binaries, no runtime deps
  • Single-shot mode — pipe-friendly for scripting

Requirements

  • Ollama running locally (ollama serve)

  • download a good model: ollama pull gemma4:e4b

Install

From source

git clone git@github.com:sha0coder/offcode.git
cd offcode
make build
make install          # → /usr/local/bin/offcode
# or
make install-user     # → ~/.local/bin/offcode

Usage

# recommended way: Interactive TUI
offcode
/model gemma4:e4b

# Single prompt (non-interactive, pipe-friendly)
offcode 'explain src/main.rs'
offcode 'write unit tests for the auth module'
offcode 'find all TODO comments in this project'

# Different model
offcode -m llama3.2 'refactor this codebase'

# Show thinking tokens (Qwen3, DeepSeek-R1, etc.)
offcode --think

# Plain terminal mode (no TUI)
offcode --no-tui

TUI key bindings

Key Action
Enter Send message
↑ / ↓ Scroll messages
PgUp / PgDn Scroll faster
← / → Move cursor in input
Home / End Jump to start/end of input
Backspace Delete char before cursor
Ctrl+C Quit

REPL commands (TUI and --no-tui)

Command Action
/help Show help
/clear Clear conversation history
/tools List available tools
/model <name> Switch Ollama model
/think Toggle thinking token display
/exit Quit

Available tools

Tool Description
read_file Read file contents (with line numbers for code)
write_file Write or overwrite a file
run_command Execute shell commands
list_dir List directory contents
search_files Search for patterns recursively
create_dir Create directories
delete_path Delete a file or empty directory
path_info File/directory metadata

Configuration

Config file is created automatically at first run:

macOS: ~/Library/Application Support/offcode/config.toml
Linux: ~/.config/offcode/config.toml

model = "gemma4:e4b"
ollama_url = "http://localhost:11434"
temperature = 0.6
num_ctx = 16384
show_thinking = false
max_tool_iters = 30
system_prompt = "You are offcode..."

View current config:

offcode --config

Cross-compilation

# Install cross (needs Docker)
cargo install cross

# Build for all platforms
make dist

# Outputs to dist/:
#   offcode-<ver>-linux-x86_64     (static, runs on any Linux)
#   offcode-<ver>-linux-arm64      (static, Raspberry Pi / Graviton)
#   offcode-<ver>-macos-x86_64    (Intel Mac)
#   offcode-<ver>-macos-arm64     (Apple Silicon)
#   offcode-<ver>-macos-universal  (fat binary, both)

Dependencies

Crate Purpose
ureq HTTP client for Ollama API (pure Rust, no tokio)
serde + serde_json JSON serialization
toml Config file parsing
dirs XDG config directory
ratatui Terminal UI

No async runtime. No OpenSSL. Fully offline after model is pulled.

License

MIT