I built a free autonomous coding agent for your terminal - bring your own API key

I got tired of paying $20/month for AI coding tools. So I built my own. Codeably is an autonomous coding agent that runs in your terminal. You bring your own API key - so it costs you $0 to run.

What it does

You describe a task in plain English. It figures out what to do and does it:

codeably "add JWT authentication to the express app"
codeably "write unit tests for every function in src/utils.js"
codeably "find and remove all dead code"
codeably "fix the race condition in payment.js"
codeably clean          # scan + remove dead code automatically
codeably review         # AI code review of your git changes
codeably chat           # open a full terminal chat UI

It runs a loop - Observe → Think → Act → Verify → Done - until the task is finished.

Try it now (no install needed)

npx codeably "your task here"

Or install globally:

npm install -g codeably

7 providers - most with free tiers

You plug in your own key. Here's what's available:

I personally use Groq for daily work - it's fast, free, and the Llama 3.3 70B model is genuinely good at code.

⚡ Type cb instead of codeably

One-command shortcut setup:

macOS / Linux / WSL:

bash install-shortcut.sh
source ~/.zshrc
cb "your task"

Windows PowerShell:

.\install-shortcut.ps1
. $PROFILE
cb "your task"

Terminal chat UI

codeably chat
# or
cb chat

Opens a full chat UI right in your terminal:

╭────────────────────────────────────────────────────────────╮
│ >/< codeably chat                                          │
│ groq · llama-3.3-70b-versatile                             │
│ type /help for commands                                    │
╰────────────────────────────────────────────────────────────╯
❯ explain how async/await works
Async/await is syntactic sugar built on top of Promises...

❯ /switch   ← switch provider mid-chat
❯ /history  ← see conversation history
❯ /save     ← save chat to file
❯ /clear    ← start fresh
❯ _

Supports markdown rendering, conversation history, and live provider switching with /switch.

30 tools

File I/O, shell commands, git operations, code search, dead code detection - 30 tools total. A few highlights:

• patch_file - safe string replacement (no full rewrites)
• grep_replace - find and replace across multiple files
• git_commit - stage all and commit with a message
• confirm_delete - always asks before deleting anything
• detect_language - auto-detects your project's stack
• run_script - write and execute multi-line bash

Safety first

Codeably cannot delete files without your explicit approval. Every deletion goes through confirm_delete:

⚠ Deletion confirmation required
Reason: unused file, zero imports detected
Files to delete:
  - src/old-utils.js
  - src/deprecated/auth.js
Delete these 2 file(s)? [y/N]:

How it's built

No frameworks. No LangChain. No magic. ~900 lines of vanilla Node.js. One OpenAI-compatible client that works with all 7 providers. A simple loop. 30 tool definitions.

codeably/
├── codeably.js       ← CLI entry + all commands
├── runtime/
│   ├── agent.js      ← Observe-Think-Act loop
│   ├── client.js     ← unified LLM client
│   └── context.js    ← codebase snapshot
├── tools/
│   └── index.js      ← all 30 tools
└── ui/
    ├── renderer.js   ← terminal UI
    └── chat.js       ← terminal chat UI

Real examples

# Add features
codeably "add rate limiting - 100 req/min per IP"
codeably "add input validation to all API endpoints"
codeably "add TypeScript types to all files in src/"

# Clean up
codeably clean
codeably "remove all console.log from src/"
codeably "migrate from callbacks to async/await in api.js"

# Tests
codeably "write integration tests for the auth routes"
codeably "add error boundary tests to the React components"

# Refactor
codeably "split the 400-line UserController into smaller files"
codeably "extract all hardcoded strings into a constants file"

# Review + docs
codeably review
codeably docs

Why I built this

Every AI coding tool I tried had the same problem - you're paying a subscription for access to a model you could call directly yourself. The tool is just a wrapper. So I wrote the wrapper. It took about a week. Now I use it daily and it costs me nothing because Groq's free tier is more than enough. The goal was always: bring your own key, own your agent.

• GitHub: https://github.com/adamyasingh-05/Codeably
• Install: npm install -g codeably

Would love feedback - what features would make this more useful for your workflow?