Key Takeaways
- Atlarix (atlarix.dev) is free β every feature, including local models and bring-your-own API keys, is on the free tier
- Core loop: plans a change, edits files, and runs the project's own tests in a real terminal; searches the codebase with plain ripgrep rather than building an index
- Every file edit is shown as a diff and only written to disk once the user approves it
- Fully local only when paired with Ollama or LM Studio β see Local vs. Hybrid below for what stays cloud-based
- A Chrome extension extends the same agent to the browser: paired with the desktop app it uses your own API keys (BYOK), same as the desktop; used on its own it runs on Atlarix's managed cloud models β a separate GitHub App, Atlarix Reviewer, always runs on those managed models β see Local vs. Hybrid below
- Atlarix Auto (free, no API key) runs on OpenAI's data-sharing program β see Platform, Pricing, and Licensing below
- Atlarix Pro is $19/month and adds Atlarix Reviewer plus included managed-inference credits
π In One Sentence
Atlarix is a free desktop AI coding agent for macOS, Windows, and Linux, built by Amariah Abishai at Norah Labs, that plans, edits, and tests code through a real terminal and only writes an edit once the user has approved it.
π¬ In Plain Terms
Atlarix is a free app you install on your computer that lets an AI agent write and test code for you, but only after you approve every change. Pair it with a local model through Ollama or LM Studio and nothing about your code leaves your machine β its Chrome extension inherits that same setup when paired with the desktop app, but runs on Atlarix's own cloud models when used on its own, same as its GitHub pull-request reviewer.
πNote: This review is based on a founder outreach email exchange with Amariah Abishai, corroborated by Atlarix's own site, plus PromptQuorum's own research into secondary sources. PromptQuorum has not independently tested Atlarix's agent or benchmarked its accuracy. See Atlarix's entry in the Local LLM Software Directory alongside other coding agents like it.
What Is Atlarix?
Atlarix is a free desktop application that runs an AI coding agent directly against a local codebase, plus a Chrome extension and a GitHub-integrated PR reviewer that extend the same agent into the browser and into pull requests. Every feature β including connecting a local model through Ollama or LM Studio and bringing your own API keys β is available without a paid tier.
- Product type: a downloadable desktop app, not a browser-only tool or an IDE plugin β plus a separate Chrome extension and a separate GitHub App
- Maker: Amariah Abishai, founder of Norah Labs
- Official site: atlarix.dev
- Cost: free tier includes every feature at no cost; Pro is $19/month β see Platform, Pricing, and Licensing below
- License and version: proprietary under an end-user license agreement, currently v14.52.1 β see Platform, Pricing, and Licensing below
What Does Atlarix Actually Do?
Atlarix's agent plans a change, edits the relevant files, and runs the project's own test suite in a real terminal β and it searches the codebase with plain ripgrep rather than building a vector index or embeddings. Every file edit is shown to the user as a diff before it touches disk; per the founder, nothing is written until that diff is approved.
- Planning and editing: the agent proposes a plan for a coding task, then edits the relevant files to implement it
- Real test execution: it runs the project's own existing tests through an actual terminal, not a simulated test runner β every command runs inside a per-OS, write-confining sandbox (Landlock on Linux, Seatbelt on macOS, a lightweight confinement on Windows) that stops it writing outside the project without changing system-wide settings
- Repo search without an index: it searches the codebase with plain ripgrep on demand, rather than pre-building an index or embeddings β a design choice the founder cites as central to the local-first guarantee
- Approval-gated writes: no file change reaches disk until the user reviews and approves the diff
- Chrome extension: the same agent runs inside the user's own signed-in browser session and can reproduce a bug and read the page's console output while fixing the underlying code β where its inference runs depends on whether it's paired with the desktop app; see Local vs. Hybrid below
- Atlarix Reviewer: a GitHub App that posts pull-request review comments (file, line, problem, suggested fix) without writing to or running the user's code
Local vs. Hybrid: The Data-Flow Breakdown
Atlarix is local when its desktop agent is connected to a local model through Ollama or LM Studio, and the Chrome extension inherits that same setup when it's paired with the desktop app. Used on its own, the extension falls back to Atlarix's managed cloud models (paid Auto) β and its Atlarix Reviewer GitHub App always runs on those managed models. The founder describes the desktop agent as "100% local with a local model, hybrid when using managed models" β this review adopts that same framing rather than a single blanket label.
Desktop agent + Ollama or LM Studio
- Where inference runs:
- Locally, on the user's own machine
- Does data leave the machine?:
- No β per the founder, no index or embeddings are built and nothing is sent out
Atlarix Auto (free, no API key)
- Where inference runs:
- Atlarix's managed cloud models, via OpenAI's data-sharing program
- Does data leave the machine?:
- Yes β free-tier requests can be used to train OpenAI's models, disclosed before the first free-tier turn
Paid Auto usage (Pro allowance or pay-as-you-go credit)
- Where inference runs:
- Atlarix's managed cloud models
- Does data leave the machine?:
- Yes, but per the founder never used for training, whether the credit comes from a Pro allowance or purchased pay-as-you-go
Chrome extension, paired with the desktop app
- Where inference runs:
- The user's own API keys (BYOK) β same as the desktop app
- Does data leave the machine?:
- No, when using a local model; yes, when the paired desktop app itself is using a cloud API key
Chrome extension, used on its own
- Where inference runs:
- Atlarix's managed cloud models (paid Auto)
- Does data leave the machine?:
- Yes
Atlarix Reviewer (GitHub App)
- Where inference runs:
- Atlarix's managed cloud models
- Does data leave the machine?:
- Yes β pull-request content is sent to Atlarix's managed models for review
A fully local, nothing-leaves-the-machine guarantee applies to the desktop agent paired with Ollama or LM Studio, and to the Chrome extension when it's paired with that same local setup. It does not apply to the extension used standalone or to Atlarix Reviewer, which are cloud components by design.
Platform, Pricing, and Licensing
Platform
- What Atlarix states:
- A free desktop app for macOS, Windows, and Linux, plus a separate Chrome extension and a GitHub App called Atlarix Reviewer.
Free tier
- What Atlarix states:
- Every feature is available for free, including local models via Ollama/LM Studio and bring-your-own API keys β no feature is gated behind a paid plan.
Atlarix Auto
- What Atlarix states:
- A free, no-API-key managed-model option. It runs on OpenAI's data-sharing program, so free-tier requests can be used to train OpenAI's models; Atlarix discloses this before a user's first free-tier turn.
Atlarix Pro
- What Atlarix states:
- $19/month. Adds Atlarix Reviewer plus included managed-inference credits. Per the founder, any paid Auto usage β from the Pro allowance or purchased pay-as-you-go credit β is never used for training.
Version
- What Atlarix states:
- v14.52.1, per the Atlarix releases page β verified 2026-09-25.
Licensing
- What Atlarix states:
- Proprietary. Free to use under an end-user license agreement (EULA) linked from the releases repository β confirmed by the founder; not open source.
πNote: The OpenAI data-sharing point is a real privacy trade-off, not a footnote: if you use Atlarix Auto's free managed-model option instead of a local model or your own API key, your requests can be used to train OpenAI's models. Atlarix says it discloses this before your first free-tier turn. If that trade-off is unacceptable, use the desktop app with a local model via Ollama or LM Studio, or bring your own API key from a provider whose terms you have already reviewed.
Get Atlarix
Atlarix (currently v14.52.1) is a free download across four channels: direct installers for macOS, Windows, and Linux, plus a Microsoft Store listing and a separate Chrome Web Store listing for the browser extension.
Source | Link |
|---|---|
| macOS, Windows, Linux installers | GitHub releases |
| Microsoft Store (signed by Microsoft β no SmartScreen warning) | Search "Atlarix" in the Microsoft Store |
| Chrome extension | Chrome Web Store |
| Official site | atlarix.dev |
Fully local use requires a local model runtime (Ollama or LM Studio); Atlarix Auto or your own API key covers cloud inference instead.
From the Maker: Amariah Abishai on Atlarix
"I built Atlarix because I wanted an AI agent I could trust with a real codebase: one that plans, edits, runs the project's own tests, and never changes a file until I approve the diff."
"With Ollama or LM Studio, everything stays on your machine; there's no index, no embeddings, and nothing leaves the laptop. The Chrome extension and Atlarix Reviewer extend the same agent to the browser and to pull requests, and those two run on our managed models."
"I'd rather state that clearly than have anyone assume more privacy than they're getting."
Quote provided directly by the founder to PromptQuorum by email and reproduced with permission, lightly trimmed for length. Amariah reviewed this article after publication and corrected several details, including the Chrome extension's data flow, the sandboxed-but-real terminal test execution, licensing, and version β see Local vs. Hybrid and Platform, Pricing, and Licensing above for the corrected facts.
Who Should Use Atlarix?
Atlarix fits developers who want a free, approval-gated coding agent that can run fully locally, and who are comfortable with a browser extension and PR reviewer that use managed cloud models instead.
Consider Atlarix if you
- Want a free AI coding agent that can run entirely locally against Ollama or LM Studio, with no paid tier required for that setup
- Want an agent that shows every file edit as a diff before it touches disk, rather than writing changes automatically
- Want the agent to run your project's real tests in a terminal, not just propose code
- Are comfortable with a PR reviewer that always runs on managed cloud models, and a browser extension that does the same unless it's paired with the local desktop app
- Want a repo-search approach that skips building a vector index or embeddings
Look elsewhere if you
- Need an open-source license before adopting a tool β Atlarix is proprietary, free under an end-user license agreement, not open source β see What Atlarix Is Not Good For below
- Want every component, including a standalone browser extension and the PR reviewer, to run on a local model β see Local vs. Hybrid above
- Are not comfortable with a free tier that can route requests through OpenAI's data-sharing program β see Platform, Pricing, and Licensing above
- Need a long public track record before adopting a coding agent β this review found no independently verifiable adoption numbers for Atlarix
What Atlarix Is Not Good For
Atlarix is not a good fit if you need an open-source license, or if you want the browser extension used standalone and the PR reviewer to run locally rather than on managed cloud models.
- Not open source β proprietary, free to use under an end-user license agreement (EULA); see Platform, Pricing, and Licensing above
- Not fully local end-to-end if the Chrome extension is used on its own β only when paired with the desktop app does it share that local/BYOK setup; Atlarix Reviewer always runs on Atlarix's managed models β see Local vs. Hybrid above for the full breakdown
- Not free of a data-sharing trade-off on its no-API-key free path β Atlarix Auto runs on OpenAI's data-sharing program, so those specific requests can train OpenAI's models
- Not independently benchmarked by PromptQuorum β this review did not measure Atlarix's coding accuracy, speed, or resource use against other agents
Common Mistakes When Evaluating Atlarix
Most confusion about Atlarix comes from treating it as a single fully local product, or from carrying over outdated claims about a competitor.
Mistake 1: Assuming the whole product is local because the desktop agent can be
The desktop agent connected to Ollama or LM Studio is local, and so is the Chrome extension when it's paired with that same desktop app β but Atlarix Reviewer, and the extension used on its own, always run on managed models. That split is by design, not a bug or a fallback β see Local vs. Hybrid above for the full per-component breakdown.
Mistake 2: Assuming Atlarix Auto never shares data
Atlarix Auto, the free no-API-key option, runs on OpenAI's data-sharing program β see Platform, Pricing, and Licensing above for what that means and how to avoid it.
Mistake 3: Assuming Continue.dev's GitHub repository is archived
Continue.dev's README says the project is "no longer actively maintained and is read-only for all users" after Cursor acquired Continue Dev, Inc. (announced June 16, 2026) and shipped its final release, v2.0.0-vscode, on June 19, 2026 β but the repository's GitHub archived flag is false; it is not technically archived. See the full Continue.dev review for the complete account.
Mistake 4: Assuming the Chrome extension always runs in the cloud
The extension only falls back to Atlarix's managed models when used on its own. Paired with the desktop app, it shares the desktop app's own setup β including a fully local one via Ollama or LM Studio β see Local vs. Hybrid above.
Competitors and Alternatives
Atlarix sits in the same segment as terminal- and IDE-native local/hybrid coding agents like Continue.dev, Cline, Aider, and Kilo Code β its main differentiator is a free tier with no feature-gating and an explicit local/hybrid split by component.
Tool | Best known for | Link |
|---|---|---|
| Continue.dev | Open-source VS Code/JetBrains assistant, now read-only after its maker was acquired by Cursor | Continue.dev review |
| Cline | Open-source, autonomous coding agent extension for VS Code and JetBrains | Cline review |
| Aider | Free, open-source, terminal-native, git-centric AI pair programmer | Aider review |
| Kilo Code | Open-source AI coding agent extension for VS Code | Kilo Code review |
This list reflects tools commonly compared in the same segment, not an independent PromptQuorum ranking β verify each tool's current pricing, licensing, and feature set before choosing.
Frequently Asked Questions
What is Atlarix?
Atlarix (atlarix.dev) is a free desktop AI coding agent for macOS, Windows, and Linux, built by Amariah Abishai, founder of Norah Labs.
Is Atlarix free?
Yes. Every feature, including local models via Ollama or LM Studio and bring-your-own API keys, is on the free tier. A $19/month Pro tier adds Atlarix Reviewer plus included managed-inference credits.
Is Atlarix fully local and private?
The desktop agent is fully local when paired with Ollama or LM Studio, and the Chrome extension shares that same setup when it's paired with the desktop app. Used on its own, the extension falls back to Atlarix's managed cloud models, and Atlarix Reviewer always runs on those managed models β see Local vs. Hybrid above for the full breakdown.
Does Atlarix Auto share my data with OpenAI?
Yes. Atlarix Auto, the free no-API-key option, runs on OpenAI's data-sharing program, so free-tier requests can be used to train OpenAI's models β see Platform, Pricing, and Licensing above.
What does Atlarix Pro cost and add?
Atlarix Pro is $19/month. It adds Atlarix Reviewer plus included managed-inference credits, and per the founder, any paid Auto usage β from the Pro allowance or purchased pay-as-you-go credit β is never used for training.
What is Atlarix Reviewer?
Atlarix Reviewer is a GitHub App that posts pull-request review comments β file, line, problem, and suggested fix β without writing to or running the user's code.
What does the Atlarix Chrome extension do?
It runs the same agent inside the user's own signed-in browser session, and can reproduce a bug and read the page's console output while fixing the underlying code.
Who built Atlarix?
Amariah Abishai, founder of Norah Labs, built Atlarix.
Has PromptQuorum independently tested Atlarix?
No. This review is a founder-submission-sourced entry, based on a founder outreach email exchange plus PromptQuorum's own corroborating research, not on hands-on testing or benchmarking of Atlarix.