3.1 · Background runs · Windows

An AI coding agent that runs on your machine.

Squig Power Code is a native IDE with an agent that plans the work, edits your code, runs your tests and checks its own results. Use a local model or your own API key. Pay once.

See it work

Squig Power Code: the agent fixed a discount bug in src/invoice.js, added a regression test, ran npm test (7 passing) and checked off its four-step task list in the chat panel.

Real run, unedited: “Invoice totals are wrong whenever a discount is applied. Find the bug, fix it, add a regression test and make sure npm test passes.”

Built for real work, not demos.

It finishes the job

The agent breaks work into a task list you can watch, edits the code, runs your tests and only reports done when the steps are checked off and the results back it up.

Review every change like a pull request

A real Source Control panel: staged and unstaged changes with line counts, inline or side-by-side diffs, commit, history, stashes and branches. Undo the agent's last edit with one shortcut.

Source Control panel listing two changed files with line counts, a commit message box and an inline diff of the one-line fix.

AI where your cursor is

Right-click any code to explain it, fix it, write tests or add docs. Ctrl+L sends a selection to chat, errors get a “Fix with agent” button, and ghost-text autocomplete uses native fill-in-the-middle on code models.

Right-click menu in the editor with Squig: Explain, Fix / Debug, Write Tests, Add Docs, Add to Chat and Edit with AI.

Autonomy with guardrails you set

Pick how much the agent may do alone. Deletes, protected files, risky commands and mass changes always ask. The rules are enforced by the app, not by the prompt, so a model can't talk its way past them.

Autonomy & Guardrails settings: Ask first, Balanced (selected) and Full autonomy, with limits for changed and deleted files per run.

New in 3.1

Background runs for local models.

Small local models do better on small steps. The pipeline splits a task into a plan and short tasks, checks each one with your tests and keeps going while you are away. In Balanced unattended mode, actions that need approval wait until you return. Best-of-N runs several attempts per task and lets the tests pick the winner.

New background task dialog with Auto, Single and Pipeline choices and the Balanced autonomy policy selected.

Start a run

Pick Auto, Single or Pipeline. Pipeline plans the work, runs each task in a fresh context, gates it on your tests, reviews it and loops on fixes.

Runs view showing a finished background run with its task status and a summary of the changes it made.

Watch it in the Runs view

See the plan and per-task status. You get a notification when a run finishes or needs you, and runs resume after a restart.

Diff review of a finished background run showing the change to src/paginate.js.

Review the result

Open the diff of what the run changed before you keep it.

Settings, Background runs: a table of model profiles and the quick model benchmark.

Tune it to your model

Model profiles set the mode per model. A quick benchmark helps you choose.

Screens show a real run on a small sample project with a local model. Results vary by model and task. Very large tasks can need several attempts, the reviewer is the same local model, and pipeline mode makes many more model calls than a single chat.

Everything else you need.

Your model, your key
Local models through Squig Model Server or LM Studio, or bring your own OpenAI, Anthropic or xAI key. Any OpenAI-compatible server works too.
Steering library
Keep your conventions in Markdown docs. A small router loads only the docs that match the request, so the context stays lean.
MCP servers
Connect Model Context Protocol servers from the MCP panel. Tool calls go through approval unless you trust them.
Helpers on big models
Capable models can hand broad searches to a read-only helper, keeping the main conversation focused.
Cost and run limits
Every turn shows its token count and cost. Per-run limits on rounds, time and spend stop runaway loops.
Native and light
Built with Rust and Tauri, not Electron. An installer under 10 MB that keeps itself up to date with signed updates.

Squig Power Code license

$49.99one-time

Use it on up to 5 devices
Automatic updates included
Local models or your own API keys
No subscription, no seat fees

Installing

The Windows installer isn’t code-signed yet, so SmartScreen may say “Windows protected your PC” with an unknown publisher. Choose More info → Run anyway.

Want to check the file first? Every release lists SHA-256 checksums. In PowerShell:

Get-FileHash .\Squig.Power.Code_*_x64-setup.exe

Updates are signed and install from inside the app. Coming from 2.x? Install 3.x once over your existing copy; your settings and license carry over.

Questions

Does my code leave my machine?: Only to the model you choose. With a local model, nothing leaves your computer. With an API key, requests go straight from the app to that provider.
Which models work best?: Anything with solid tool calling. Our live test suite runs the agent against Qwen3 on a local server, and Claude, GPT and Grok models are supported through your own key.
Is there a subscription?: No. One payment covers the license on up to 5 devices, and updates install automatically.
Mac or Linux?: Version 3 is Windows 10/11 (x64). Linux builds are next.

Looking for the other apps? See all products.