quincode

quincode is a Rust terminal coding agent with:

• Interactive TUI mode (quincode)
• One-shot mode (quincode run "<prompt>")
• Config doctor (quincode config doctor)
• OpenAI-compatible and Anthropic-compatible API support
• Shell tool execution with policy + approval controls

Current Status

Implemented in this repo today:

• Local/project config layering from ~/.quincode/config.toml and <repo>/.quincode/config.toml
• Interactive terminal UI with slash commands, command history, multiline input
• shell_command tool loop with timeout, allow/block rules, and approval modes
• Provider adapters for OpenAI-style and Anthropic-style APIs

Not yet implemented from the original plan:

• Session persistence and transcript export commands
• Extra tools (read_file, write_file, list_files, search_text)

Build and Run

Prerequisites:

• Rust toolchain (stable)

Commands:

# Build
cargo build

# Run interactive mode
cargo run

# Run one-shot mode
cargo run -- run "inspect this repository"

# Validate config + endpoint/model visibility
cargo run -- config doctor

# Run tests
cargo test

CLI Commands

quincode
quincode run "<prompt>"
quincode config doctor

Interactive Controls

Input and navigation:

• Enter: submit
• Shift+Enter: newline
• Left/Right: move cursor
• Up/Down: command history (or slash suggestion selection when input starts with /)
• PageUp/PageDown: transcript scroll
• Esc: quit
• Ctrl+C twice: quit confirmation
• Ctrl+O: expand last truncated tool output

Slash commands:

• /help
• /expand
• /model
• /model <id>
• /quit

Configuration

Config file precedence:

1. ~/.quincode/config.toml
2. <repo>/.quincode/config.toml (overrides global)

Current schema:

[api]
compatibility_mode = "openai" # openai | anthropic
base_url = "http://localhost:1234/v1"
model = "qwen2.5-coder-32b-instruct"
api_key = ""

[agent]
max_steps = 4
approval_mode = "never" # never | on-request | always

[tools]
shell_enabled = true
shell_timeout_secs = 120
allowed_commands = []
blocked_commands = ["rm -rf /", "mkfs", "shutdown"]

Notes:

• .quincode/config.toml is gitignored for local secrets and endpoint settings.
• allowed_commands = [] means no allowlist restriction (only blocklist applies).
• In on-request mode, risky commands require confirmation.

Provider setup guides:

• Anthropic setup
• OpenAI setup
• Local OpenAI-compatible setup (Ollama/LM Studio/vLLM/LocalAI)

Recent Changes (from git log)

• 60dd977: hardened terminal resize and wrapping behavior
• 794ef39: cleaner inline prompt collapse on quit
• 808c878: clear inline viewport on quit
• a98848d: up/down command history in input
• 1930090: inject cwd + local AGENTS.md into runtime system prompt

Project Layout

src/
  main.rs            # CLI entrypoint + command dispatch
  cli.rs             # clap command definitions
  config.rs          # config schema + layering
  agent.rs           # agent loop + tool orchestration
  prompt.rs          # system prompt + local AGENTS.md injection
  llm/mod.rs         # provider adapters (openai/anthropic)
  tools/shell.rs     # shell tool runner
  ui/                # ratatui app/input/formatting/terminal loop