Frequently Asked Questions

Attune AI is a spec-driven development platform that combines AI workflows, project memory, retrieval grounding, and verification tools to help teams turn requirements into reliable software. It runs on a Claude subscription or an API key, and is open source under Apache 2.0.

The reliability loop is the arc Attune AI takes a requirement through on its way to shipped software: Specify, Ground, Build, Remember, Verify. You write a spec, ground every change in your real code, build with multi-stage workflows, carry what worked into the next session, and verify the output before it ships.

Attune AI is built on four pillars, each a real, shipped capability: AI workflows (19 multi-stage workflows for review, tests, bug prediction, and refactors), Project memory (cross-session findings and a retrievable lessons corpus), Retrieval grounding (citations back to your source via attune-rag), and Verification (fact-checking generated content before it reaches main).

Attune AI ships 22 workflows (19 multi-stage), 47 MCP tools, 23 auto-triggering skills, 5 wizards, and 15 template kinds. The spec engine runs via /spec, progressive help via /coach, and cross-session recall via /recall.

Two packages today, one workflow. attune-ai is the framework — workflows, agents, the CLI. attune-gui is the local dashboard that sits on top of it: Living Docs, RAG search, the commands surface, summaries. You install them separately right now (pip install attune-ai and pip install attune-gui) and they talk to the same project directory. In an upcoming release they fold into one install (pip install attune-ai[gui]); see the migration page for the heads-up.

No — the CLI does everything on its own. But the dashboard is where the workflow tells you what to do next: which docs have drifted, which summaries need re-polishing, which commands you ran recently and what they produced. If you’re just running one-off workflows, the CLI is fine. If you’re maintaining a living help system or iterating on a corpus, the dashboard saves real time. Migration heads-up here if you have attune-gui installed today.

Attune AI is completely free and open source under the Apache License 2.0. Use it in personal projects, startups, or large enterprises at no cost. No license keys, no restrictions, no hidden fees.

Apache 2.0 is a permissive open source license approved by the OSI. It allows you to use, modify, distribute, and sell products built with Attune AI. It includes patent protection and is approved by most enterprise legal teams.

Yes! Apache 2.0 explicitly permits commercial use. Build and sell products using Attune AI without any licensing fees or restrictions.

Retrieval grounding is powered by attune-rag, a core (built-in) dependency. Keyword and semantic retrieval keep generated content anchored to your actual source, with citations back to where each claim came from. Mean faithfulness is measured at ≥ 0.97 and CI-gated — drift fails the build rather than slipping through.

Verification fact-checks LLM output against your source-of-truth before a change reaches main: it confirms imports actually import, CLI flags are real, links resolve, and counts match. It closes the loop the spec opened, so generated docs and code stay honest.

Memory is local-first — nothing is sent to a cloud by default. Findings and the lessons corpus live on your machine. An optional Redis semantic tier is available for richer recall, and it uses local Ollama embeddings, so you stay in control of where your data goes.

No. Attune AI works on a Claude subscription or an API key — use whichever you have. The platform is open source under Apache 2.0 with no license keys or hidden fees.

Attune AI is built for Anthropic Claude with a Claude-native architecture, so it leverages Anthropic's automatic prompt caching, extended thinking, and optimized tool use.

The platform is cross-platform and runs on macOS, Linux, and Windows. It requires Python 3.10+ and works with all major development environments including VS Code, JetBrains IDEs, and terminal-based workflows.

Wizards are guided, multi-step AI workflows that walk you through complex tasks. Each wizard collects context via questions, runs AI analysis, decomposes work into tasks, and previews results before acting. Attune AI ships with 5 built-in wizards.

From Claude Code, type /wizard run debug (or any wizard ID). From Python: from attune.wizards import get_wizard; wizard = get_wizard("debug")(); result = await wizard.run(). See the Getting Started guide for a full walkthrough.

Yes! Two approaches: (1) YAML-based — create a .attune/wizards/my-wizard.yaml file with step definitions, no Python required. (2) Python-based — subclass BaseWizard for advanced logic like workflow delegation, conditional steps, and custom result processing. See the Custom Wizard Development guide.

You take requirements through to reliable software: write a spec with /spec, ground changes in your real code, run multi-stage workflows for code review, security scanning, test generation, bug prediction, and refactor planning, then verify the output before it ships. The lessons corpus and cross-session memory carry what worked into the next session.

The spec engine runs via /spec. It guides you Socratically through requirements, design, and tasks with an approval gate, so a feature starts from an agreed spec rather than an open-ended prompt — the first stage of the reliability loop.

As of January 28, 2026, we switched from Fair Source 0.9 to Apache 2.0. We realized the licensing restrictions were limiting adoption without generating revenue. Going fully open source lets us focus on building the best platform and growing a community.

Get community support via GitHub Discussions. Report bugs via GitHub Issues. Enterprise users can reach out for dedicated support options.

Report bugs via GitHub Issues. Include your environment details, steps to reproduce, and expected vs actual behavior.

Yes! We welcome contributions. Check out our GitHub repository for contribution guidelines. The framework is fully open source under Apache 2.0, making it easy to fork, modify, and contribute back.