SparrowCodes CLI

A coding agent that picks the right model for each task

The SparrowCodes CLI works in your terminal. It reads your project, edits files and runs your commands and tests. Sparrow Auto sends routine work to fast, low-cost models and hard work to stronger ones, so you don't pay frontier prices for renaming a variable.

$ curl -fsSL https://sparrowcodes.com/install | sh

Free plan: 50 requests a day, no card needed
macOS, Linux and Windows 10 or newer. No sudo or admin rights
Every request in your activity log, with its exact cost

Start free Read the docs

$ sparrowcodes
> Write tests for invoice.py and fix any bugs they find

  Sparrow Auto picked the model for this task
  ✓ Wrote tests/test_invoice.py, 14 tests
  $ pytest -q
  3 failed, 11 passed
  2 bugs: discount divides by 10, split loses a cent
  ✓ Fixed apply_discount() and split_bill()
  $ pytest -q
  14 passed in 0.01s

An example session, shortened.

What it does

Give it a task. Check the diff.

Describe what you want in plain words. The CLI looks around your code, makes a plan, changes the files and checks its own work by running your tests.

Reads your whole project: It searches and opens the files it needs, so you don't paste code into a chat window.
Edits and writes files: Changes show up as normal edits in your working tree. Review them with git like any other change.
Runs commands and tests: It runs your build and tests, reads the failures and fixes them, then runs them again.
Asks before it acts: Shell commands, file edits, anything outside the project folder and web fetches all need your OK.
Any model in the catalog: Use Sparrow Auto, or pick a model with -m: Claude, GPT, Gemini, DeepSeek, Qwen and more.
Scripts and CI: Run one task with sparrowcodes run and set SPARROWCODES_API_KEY instead of logging in.

Sparrow Auto

Strong models when it matters, low-cost ones when it doesn't

Auto picks a model once per task, based on what the task needs. If a task keeps failing, it moves to a stronger model instead of retrying the same one. It stays within your plan, and on the free plan it uses the free model.

In our test of ten coding tasks, Auto passed everything Claude Sonnet 5.5 passed, for 92% less
The model it chose is shown in the sidebar for every task
Pick a model yourself any time with -m

How Auto chooses

$ export SPARROWCODES_API_KEY="sk-sc-v1-…"
$ sparrowcodes run "explain this repo"

# Approves every action. Only in a folder you trust.
$ sparrowcodes run --auto "add a --verbose flag to the CLI"

# Pin a model instead of Auto
$ sparrowcodes run -m sparrowcodes/anthropic/claude-sonnet-5.5 "review the diff"

sparrowcodes run cannot ask for permission, so it refuses edits and commands unless you add --auto.

Commands

Seven commands to learn

That's the whole list. Everything else happens in the conversation.

Full reference

sparrowcodes: Start an interactive session in the current folder.
sparrowcodes run "…": Do one task, then exit. Good for scripts.
sparrowcodes models: List the models you can use.
sparrowcodes run -m <model> "…": Use a specific model instead of Auto.
sparrowcodes auth login: Paste your API key once. It stays on your computer.
sparrowcodes upgrade: Update to the newest version.
sparrowcodes uninstall: Remove the CLI and its folder.

Pricing

Free to start, flat price when you use it every day

Plans last 30 days and never renew by themselves. Without a plan, the CLI spends from your credits at the same list prices as the API.

Free: 50 requests a day on the free model
Developer: Every model, including Claude Sonnet and Opus
Pro: Every model, including Claude Sonnet and Opus
Credits: Pay per request, any model

Compare plans

Already use another coding tool?

SparrowCodes also works as the provider behind Claude Code, Cline, Continue, Aider and Cursor.

Install the CLI in one line

Start free Read the docs