https://www.atlarix.dev/blogs How Atlarix, the AI coding agent by NorahLabs, works, and how it compares. en https://www.atlarix.dev/blogs/atlarix-auto-free-ai-coding-agent-no-api-key https://www.atlarix.dev/blogs/atlarix-auto-free-ai-coding-agent-no-api-key Atlarix Auto is a free AI coding agent mode with no API key: it picks a model per step (explore, build, plan, images), is free in the desktop app via OpenAI's data-sharing program, and private and never trained on with Pro or credit. Tue, 01 Sep 2026 00:00:00 GMT https://www.atlarix.dev/blogs/atlarix-vs-claude-code-pricing-and-platforms https://www.atlarix.dev/blogs/atlarix-vs-claude-code-pricing-and-platforms Atlarix vs Claude Code compared on packaging: Code Review averages $15-25 per review on Team and Enterprise only, Chrome integration needs a direct Anthropic plan. Atlarix includes reviewer and extension in Pro. Tue, 01 Sep 2026 00:00:00 GMT https://www.atlarix.dev/blogs/atlarix-vs-codex-local-models-and-licensing https://www.atlarix.dev/blogs/atlarix-vs-codex-local-models-and-licensing Atlarix vs OpenAI Codex: both run local models, so that is a tie. The differences are the approval queue, the per-OS write sandbox, review metering, and an Apache-2.0 CLI versus a proprietary workstation. Tue, 01 Sep 2026 00:00:00 GMT https://www.atlarix.dev/blogs/atlarix-vs-cursor-how-each-finds-code https://www.atlarix.dev/blogs/atlarix-vs-cursor-how-each-finds-code Atlarix vs Cursor, fact-checked: Cursor's docs now describe Instant Grep with no mention of embeddings, so we cut that contrast. What remains: where search runs, how Bugbot is billed, and editor versus workstation. Tue, 01 Sep 2026 00:00:00 GMT https://www.atlarix.dev/blogs/atlarix-vs-github-copilot-indexing-and-privacy https://www.atlarix.dev/blogs/atlarix-vs-github-copilot-indexing-and-privacy Atlarix vs GitHub Copilot: Copilot builds a semantic code search index and runs review on GitHub Actions runners with two cost components. Atlarix indexes nothing, runs no runner, and bills one balance. Tue, 01 Sep 2026 00:00:00 GMT https://www.atlarix.dev/blogs/atlarix-vs-opencode-permissions-and-review https://www.atlarix.dev/blogs/atlarix-vs-opencode-permissions-and-review Atlarix vs opencode: opencode starts from permissive defaults and runs review inside your own GitHub Actions runner. Atlarix asks before every write and runs a reviewer with three read-only tools and no runner. Tue, 01 Sep 2026 00:00:00 GMT https://www.atlarix.dev/blogs/benchmarking-a-coding-agent-honestly https://www.atlarix.dev/blogs/benchmarking-a-coding-agent-honestly An honest Terminal-Bench 2.0 write-up: 42 of 88 scored tasks versus 39, at single attempt, with a quarter of trials timing out. Why it proves nothing, the denominator error we found on our own page, and why no SWE-bench figure exists. Tue, 01 Sep 2026 00:00:00 GMT https://www.atlarix.dev/blogs/coding-agents-that-dont-index-your-code https://www.atlarix.dev/blogs/coding-agents-that-dont-index-your-code Why some coding agents build no index: what an embedding index commits you to, what lexical search gives up, and why Atlarix deleted its own FTS5 and BM25 stack after shipping it. The trade, stated both ways. Tue, 01 Sep 2026 00:00:00 GMT https://www.atlarix.dev/blogs/how-we-deleted-our-retrieval-index https://www.atlarix.dev/blogs/how-we-deleted-our-retrieval-index Why Atlarix deleted its FTS5, BM25, ctags and reranker retrieval stack in v14.9.0: a chokidar watcher held one fd per file, hit roughly 12,365 on a large repo, and caused spawn EBADF. Net 5,246 lines removed. Tue, 01 Sep 2026 00:00:00 GMT https://www.atlarix.dev/blogs/private-workstation-ai-coding-agent https://www.atlarix.dev/blogs/private-workstation-ai-coding-agent AI coding agent privacy, mechanically: no index and no embeddings means nothing about your code is stored anywhere. What does leave the machine, what the approval queue and OS sandbox cover, and the reviewer that never executes code. Tue, 01 Sep 2026 00:00:00 GMT https://www.atlarix.dev/blogs/run-a-coding-agent-on-local-models https://www.atlarix.dev/blogs/run-a-coding-agent-on-local-models How to run a coding agent entirely on local models with Ollama or LM Studio: connecting the endpoint, picking a tool-capable model, why no-index retrieval helps on small context windows, and where local still falls short. Tue, 01 Sep 2026 00:00:00 GMT https://www.atlarix.dev/blogs/stopping-a-turn-should-not-cost-more-than-finishing-it https://www.atlarix.dev/blogs/stopping-a-turn-should-not-cost-more-than-finishing-it Atlarix v14.45 fixes cancellation billing: a stopped generation is now priced from what was really generated and can never exceed the estimate you were quoted. The same request measured 45c completed and 58c cancelled before the fix. Tue, 01 Sep 2026 00:00:00 GMT https://www.atlarix.dev/blogs/what-atlarix-is-and-is-not https://www.atlarix.dev/blogs/what-atlarix-is-and-is-not The canonical description of Atlarix, with every outdated and incorrect claim about it corrected and dated: Blueprint, tool counts, plan gating, the Team tier, Windows, and the open-source reviewer claim we retracted. Tue, 01 Sep 2026 00:00:00 GMT https://www.atlarix.dev/blogs/write-confining-sandboxes-on-three-operating-systems https://www.atlarix.dev/blogs/write-confining-sandboxes-on-three-operating-systems How Atlarix confines agent command writes on macOS, Linux and Windows: Seatbelt, Landlock with a bubblewrap fallback, and Low Integrity Level after AppContainer broke the Windows toolchain. Including three limits of the sandbox. Tue, 01 Sep 2026 00:00:00 GMT https://www.atlarix.dev/blogs/how-atlarix-uses-your-codebase-as-a-rag-knowledge-base https://www.atlarix.dev/blogs/how-atlarix-uses-your-codebase-as-a-rag-knowledge-base Atlarix gives AI full codebase context without RAG or embeddings: bundled ripgrep (grep + glob) for fast lexical search and the model's own reasoning. No index, no vector store, nothing uploaded — by design. Mon, 01 Jun 2026 00:00:00 GMT https://www.atlarix.dev/blogs/how-to-evaluate-a-coding-agent https://www.atlarix.dev/blogs/how-to-evaluate-a-coding-agent A practical guide to evaluating coding agents in 2026: model flexibility, retrieval approach, approval model, data path, offline support, and extensibility. No product names — just the criteria that matter. Mon, 01 Jun 2026 00:00:00 GMT https://www.atlarix.dev/blogs/local-llm-coding https://www.atlarix.dev/blogs/local-llm-coding Run the open-weight frontier locally via Ollama and LM Studio with no-embeddings retrieval that never leaves your machine. Native tool history and an XML/text tool fallback are why local models actually perform here. Mon, 01 Jun 2026 00:00:00 GMT https://www.atlarix.dev/blogs/why-atlarix-is-a-desktop-app-not-a-vscode-extension https://www.atlarix.dev/blogs/why-atlarix-is-a-desktop-app-not-a-vscode-extension Why Atlarix is a desktop app, not an editor extension: a real runtime for local models, a full approval queue with a per-OS write sandbox, an in-app browser and terminal, and a local-by-design data path — beside any editor. Mon, 01 Jun 2026 00:00:00 GMT https://www.atlarix.dev/blogs/blueprint-context-management-research https://www.atlarix.dev/blogs/blueprint-context-management-research Atlarix Blueprint research published on Zenodo: section-scoped structural graph RAG delivers ~6,500 tokens of codebase understanding. Benchmark results, honest findings, and what the counterintuitive outcome means for context management. Fri, 01 May 2026 00:00:00 GMT