ainacode (aina)

An AI coding assistant for your terminal. aina can read and edit files, run commands, search your codebase, and complete multi-step tasks agentically — right from your shell, with colored diffs, change confirmations, file autocomplete, and a built-in bash mode.

It connects to the api.tiarina.id gateway using first-party models (aina-1-flash / aina-1-mini / aina-1-pro / aina-1-ultra).

Features

• 🤖 Agentic tool-use — reads, writes, and edits files, runs commands, searches code (regex grep + glob find), and inspects git.
• 🩹 Robust edits — exact match with a whitespace-tolerant fallback; apply many edits at once, atomically.
• 🔁 Self-checking — after changing files, aina can auto-run your typecheck/lint and fix the errors it introduced.
• 📋 Default / Plan / Auto modes — Plan is read-only (drafts a plan only), Auto skips confirmations. Press Tab to cycle.
• ✅ Safe by default — file changes and commands show a diff/preview and ask for approval. Actions outside the working folder always require confirmation, and risky commands (e.g. rm -rf, sudo) always need explicit confirmation.
• 🧠 Project context — automatically reads AINA.md / CLAUDE.md / AGENT.md from your working folder; /init generates an AINA.md for you.
• ↩️ Undo — revert the last file change with /undo.
• 📎 File attach @ — type @path to drop a file's contents into your message.
• 🐚 Bash mode ! — run a shell command directly without the model.
• 📨 Message queue — keep typing while a task runs; queued messages run next.

Requirements

• Node.js ≥ 18

Installation

npm install -g ainacode

Configuration

The first time you run aina without an API key, it will prompt for one and save it to ~/.ainacode/config.json.

You can also configure it via environment variables (these take precedence over the config file):

Variable	Description	Default
AINA_PROVIDER	Provider id (tiarina, openai, openrouter, or custom id)	—
AINA_API_KEY	Provider API key	—
AINA_MODEL	Default model	provider default
AINA_BASE_URL	Base URL for custom OpenAI-compatible providers	preset provider URL
AINA_AUTO_VALIDATE	Auto typecheck/lint after edits (false/0 disables)	true
AINA_VALIDATE_CMD	Override the validation command (e.g. npm run typecheck)	auto-detected

On first run, AINA opens an interactive provider setup wizard. Tiarina API is recommended and only requires an API key. OpenAI and OpenRouter also only need an API key; custom OpenAI-compatible providers ask for name, URL, and key. You can manage providers later with /provider and view settings with /config. Provider setup validates the connection by fetching models first. First-run uses the first fetched model automatically; adding a provider later saves it without switching away from your active provider.

Usage

aina                                   # interactive shell (REPL)
aina "explain src/index.ts"            # one-shot query
aina -m aina-1-pro "optimize this function"   # override the model
aina -y "tidy up all imports"          # start in auto-approve mode
aina --version                         # print the version
aina --help                            # print help

CLI options

Option	Description
-m, --model <model>	Use a model for this invocation
-y, --yes	Auto-approve mode (skip tool confirmations)
-v, --version	Print the version
-h, --help	Print help

Inside the REPL

Command	Action
/help	Show help
/provider	Manage OpenAI-compatible providers
/model [name]	Show or switch the active model
/init	Generate/update AINA.md (project context)
/undo	Undo the last file change (write/edit/delete/move)
/check	Run validation (typecheck/lint) now
/diff	Show the git diff of your changes
/commit "message"	Stage all & commit (with confirmation)
/resume	Resume your previous session
/compact	Summarize & shrink the conversation context
/usage	Show input/output token usage, cost estimate & context size
/status	Show model, mode, directory, session & usage
/config [key value]	Show or change configuration
/plan	Toggle Plan mode (read-only, plan only)
/auto	Toggle auto-approve mode
/clear	Reset history & clear the screen
/exit	Quit
Tab	Switch mode (Default → Plan → Auto)
@<file>	Attach a file's contents to your message
!<command>	Run a bash command directly

Press Esc to cancel a running task.

Providers & models

AINA supports OpenAI-compatible providers:

Provider	Setup	Default/preset models
Tiarina API (recommended)	API key only	aina-1-flash, aina-1-mini
OpenAI	API key only	fetched from OpenAI, fallback gpt-5.5
OpenRouter	API key only	fetched from OpenRouter, fallback openai/gpt-5.5
OpenAI Compatible	name, base URL, API key	fetched from provider

The /provider manager validates and caches model lists before saving new keys. The /model picker shows provider headers with indented model choices, so you can select a visible model directly or open More Models for a provider. Type in the outer picker to search across providers; type inside provider details to search only that provider. Manual model IDs are still supported.

If a provider emits explicit <thought> or <think> text, AINA shows a short dimmed thinking preview above the answer and hides the raw tags from the reply.

Tiarina models

ID	Name
aina-1-flash	Aina 1 Flash (default, fast & balanced)
aina-1-mini	Aina 1 Mini (lightweight & very fast)
aina-1-pro	Aina 1 Pro (advanced reasoning & coding)
aina-1-ultra	Aina 1 Ultra (maximum intelligence)

Modes & safety

• Default — asks for confirmation before changing files or running commands.
• Plan — read-only; investigates and drafts a plan, changes nothing.
• Auto — skips confirmations for a faster session.

In every mode, actions targeting files outside the working folder always ask for confirmation, and risky commands (rm -rf, sudo, mkfs, git push --force, piping to a shell, …) always require explicit confirmation. Code search and the file picker respect your .gitignore.

Privacy

• Provider API keys are read from AINA_API_KEY or ~/.ainacode/config.json (stored locally). Without a configured provider, aina opens setup or stops with a clear message in non-interactive mode.
• Files you read or @-attach, and commands you run, are sent to the active provider as part of the context. Respecting .gitignore helps avoid sending secrets (e.g. .env) by accident.
• Your resumable session history is stored locally under ~/.ainacode/sessions/<uuid>.json, with the latest pointer at ~/.ainacode/last-session.json.
• /usage shows prompt/completion/total tokens plus a best-effort cost estimate for known Aina models; unknown model prices are shown as unavailable.
• diag.mjs requires a real API key from AINA_API_KEY or config and exits clearly when none is configured.

Platform Notes

• Linux and macOS terminals are the primary supported environments.
• Windows dangerous-command detection is included. For best results, use Windows Terminal with PowerShell and Node.js 18+.
• Full raw TTY/ANSI behavior is best-effort on Windows and has not been fully verified in this audit, so avoid claiming full Windows support until tested on native Windows terminals.

Quality Gate

• CI runs npm run build, npm test, and npm run lint on push and pull requests.

License

ISC — see LICENSE.