Get This Tool
Collie
Pricing
- Model
- Free
Summary
Most coding agents give you advice. They suggest the fix, sketch the diff, explain the approach — then hand it back to you to actually run it. Collie runs it, checks it passed, and only tells you it's done when it is.
Collie is a local, open-source coding agent that operates directly on your machine — your file system, your signed-in browser session, your real terminal. The core loop is: you describe the task in plain language, Collie does the actual work, then writes a verification step and runs it before declaring the job complete. That proof-first model is what separates it from a chat assistant. It installs as a desktop app on Windows and macOS, or via a single pip command on Linux. No telemetry, no cloud relay — your files and credentials stay local.
Bottom line: Pick Collie when you need a local agent that can hit your real browser session and verify its own fixes — but if your work requires calling external APIs, triggering webhooks, or integrating with CI pipelines, you will be wiring those connections yourself.
Community Performance Report Card
No community ratings yet. Be the first to rate this tool!
Pros
Sign in to edit- Runs against your real browser session and signed-in accounts, so tasks like pulling data from a portal or filling a form work without re-authentication or credential sharing with a third party.
- Writes and runs a verification step before calling a task complete, which means you get a passing test as proof of a bug fix rather than a diff you have to validate yourself.
- No telemetry and no cloud relay — your source code, files, and session credentials stay on your machine, so it fits workflows where data cannot leave the local environment.
- One-click desktop install on Windows and macOS with no admin rights required, so setup does not block a developer who lacks elevated permissions on a managed machine.
- MIT-licensed and self-hosted, so you can inspect the source, fork it, or audit what it does — which matters when you are handing an agent access to your file system and browser.
Cons
Sign in to edit- No API surface exists, so you cannot trigger Collie from a script, a CI job, or an external scheduler — any team that needs to embed AI task execution inside an automated pipeline will have to run it interactively or switch to an agent framework that exposes a callable interface.
- Linux installation requires Python 3.12 or later and a pip install from GitHub; teams on managed Linux environments where Python version is locked by policy will need to resolve that dependency before anything runs.
- The tool is scoped to single-session, single-machine operation with no documented multi-agent coordination — workflows that need parallel agents handing off between steps are outside what the current architecture supports, and teams building those patterns will move to a framework designed for it.
About
- Platforms
- Windows, macOS, Linux
- API Available
- No
- Self-Hosted
- Yes
- Last Updated
- 2026-08-14T03:04:16.565Z
Best For
Who it's for
- Developers needing local AI assistance with real environment access
- Users wanting private, no-telemetry task automation
- Workflows that require browser control and file system changes
What it does well
- Fix bugs and verify fixes by running reproduction tests
- Automate browser-based tasks while signed into accounts
- Organize files and summarize documents on the local machine
- Set up or migrate projects and run setup commands
Add notes, reviews, and benchmarks so the next visitor gets a clearer picture.
Compare Collie
Spotted incorrect or missing data? Join our community of contributors.
Sign Up to ContributeFrequently Asked Questions
- Is Collie free?
- Yes — Collie is fully free to use. There is no paid tier.
- Is Collie open source?
- Yes. Collie is open source.
- Can I self-host Collie?
- Yes. Collie supports self-hosting on your own infrastructure.
- What platforms does Collie support?
- Collie is available on: Windows, macOS, Linux.
Curated lists that include this category
Coding agents that run in sandboxed environments look great until the task requires something real — a browser logged into your staging account, files on your desktop, a test suite that only passes against your local database. Collie is built around that gap. It installs on your machine and operates against your actual environment: opening your browser in your existing session, editing your files, running your terminal commands. The stated workflow is: describe the task in plain language, Collie executes each step, writes a reproduction check, runs it, and hands back the result only when it passes. The checkout-bug example the vendor documents — find the line, write a failing test, apply the fix, re-run to green — illustrates the intended loop.
The differentiating feature is the verification step. Rather than applying a change and stopping, the vendor states Collie writes a check and runs it before marking the task done. For bug fixes this means a reproduction test that was red before the patch and green after. For browser tasks it means confirming the action completed in your actual session. This shifts the output from ‘here is a diff’ to ‘this passed’.
Collie fits solo developers and small teams who want local task automation without sending code or credentials to a hosted service. The MIT license and self-hosted-only architecture mean no data leaves the machine. Where it breaks: there is no API surface, so you cannot slot Collie into an automated pipeline or trigger it from a CI step. There is no mention in the docs of multi-agent coordination, so tasks that require agents working in parallel are outside its scope. Teams who need to embed AI task execution inside an existing workflow — rather than running it interactively — will hit that wall quickly and look elsewhere.
Installation is a desktop app on Windows (no admin rights required) and macOS (Apple-signed, drag-to-install), or pip from the GitHub repository on Linux with Python 3.12 or later required. The tool runs on your own AI subscription — the vendor states Collie uses your existing API key rather than proxying through a hosted service. Version v0.20.24 is the release documented on the page, published under the MIT license.