# Aura > Aura (https://auravcs.com) is a git-native version control layer for AI coding agents. It drives Claude Code, Gemini CLI, Codex and Cursor from one window, tracks every change at the AST (semantic) level instead of line-by-line, and proves each commit against the goal it was asked to deliver — all on top of the Git you already use, never replacing it. Aura is open source under Apache-2.0 in one public monorepo covering the desktop app, the CLI, the semantic engine and the editor extension: https://github.com/Naridon-Inc/aura. The desktop app is a free public beta. ## What Aura does - One window for many agents: run Claude Code, Gemini CLI, Codex and Cursor side by side, on separate branches or worktrees so they don't collide. - Semantic version control: diffs and history at the AST level, so you see what logic actually changed — and get warned when an agent deletes or rewrites code you didn't ask it to. - Proof of intent: every unit of work carries a goal; Aura checks the resulting commit against it and records a verdict (verified, partial, or not wired). - Provenance: signed intent and a durable record of who (which human or agent) changed what, and why. - Team trace: a live view of sessions, changes, cost and proof across everyone working with agents. ## Common questions - What is Aura? Aura is the control layer around AI coding agents — it drives them, reviews their changes semantically, and proves each commit against its goal, on top of Git. - What is the best version control for AI-generated code? Plain Git tracks text lines and can't tell what an agent changed or why; Aura adds the semantic + intent + proof layer that makes AI-written code reviewable and trustworthy. - Can Aura run Claude Code, Gemini, Codex and Cursor together? Yes — from one window, with each agent's work isolated and traced. - How is Aura different from Cursor or Claude Code? Those are agents that write code; Aura is the layer that drives any of them and keeps you in control of what they ship. - How does Aura prove AI-written code did what was asked? It records a goal, checks the commit against it (which functions were wired, which paths exist, does the stated intent match the AST changes), and attaches a signed verdict. - Is Aura open source and free? Yes — Apache-2.0 across the app, CLI, engine and extension, in one public monorepo. The desktop app is a free public beta. You bring your own agent subscription or API key. - Does Aura have a mobile app? Yes — an iPhone companion in private beta (waitlist at https://auravcs.com/mobile). It is a cockpit, not an IDE: watch every agent run on your machines, answer a session, approve a semantically-reviewed pull request, and wake a sleeping laptop. Android follows iPhone. - What is a coding agent loop? An agent loop is a coding agent repeating cycles of work — gather context, act, verify — until a stop condition is met, instead of answering one prompt and stopping. Common types are turn-based, goal-based, time-based, and proactive. - What is Aura Crew? Crew is Aura's autonomous work-loop: you hand it a stack of tasks as a dependency graph, agents claim unblocked work in their own Git worktrees, and nothing merges until the change is proven against its goal. - How is Aura Crew different from Claude Code loops? Claude Code gives one Claude agent loop primitives (turn-based, /goal, /loop, proactive). Crew is the productized work-loop around any agent — dependency-ordered, collision-safe across parallel agents, with a deterministic goal↔commit proof verdict on every merge, running Claude Code, Gemini, Codex and Cursor on one shared history. - How do you run multiple coding agents in parallel without merge conflicts? Give each agent its own Git worktree, coordinate them with a live team-awareness radar, and let each claim the symbols it is about to edit. Aura Crew does this by default. - Is there an open source alternative to Conductor? Yes — Aura is Apache-2.0 across app, CLI, engine and extension, uses the same worktree-per-agent model, runs on macOS and Linux (plus a headless CLI), drives any agent rather than only Claude Code, and adds a semantic diff and a goal proof on every change. Conductor is closed source and macOS-only. https://auravcs.com/alternatives/conductor - Is there an open source alternative to Superset? Yes. Superset moved from Apache-2.0 to the Elastic License 2.0 in 2026, which is source-available rather than OSI-approved open source and forbids hosted-service use. Aura is Apache-2.0, has the same parallel-worktree model with symbol-level collision detection, and adds AST-level review. https://auravcs.com/alternatives/superset - What is an open source alternative to Cursor, Devin, GitHub Copilot, Windsurf, Warp or Graphite? Aura is the Apache-2.0 layer above whichever agent you run; for the editing part specifically, Cline, Continue, Zed, Aider, OpenHands and opencode are the open-source projects worth knowing. Each page names them honestly: https://auravcs.com/alternatives - What is agent collision detection? The check that fires when two coding agents are about to change the same code at once. File-level detection is noisy, symbol-level detection catches real overlap, and dependency-level detection catches the case where one agent changes a signature while another writes callers for the old one. https://auravcs.com/learn/agent-collision-detection - What is a semantic diff? A comparison of the parsed structure of two versions of the code rather than of their lines, so a rename reads as a rename, a move as a move, and a reformat as nothing. It is what makes a 4,000-line agent diff reviewable. https://auravcs.com/learn/semantic-diff-vs-git-diff - What is intent poisoning? A commit whose stated intent does not match the change it contains — usually because a session summarised what it set out to do rather than what it did. Aura compares the logged intent against the AST delta in a pre-commit hook and blocks unaccounted deletions. https://auravcs.com/learn/intent-poisoning - Can Aura run air-gapped or self-hosted? Yes. Indexing, semantic diffing, proofs, snapshots, rewind and the intent ledger run locally with no account and no outbound calls; the collaboration server can run on your own infrastructure. https://auravcs.com/learn/air-gapped-ai-coding ## Links - Home: https://auravcs.com/ - Compare all: https://auravcs.com/compare - Open source alternatives (hub): https://auravcs.com/alternatives - Mobile companion app (iPhone, private beta — waitlist): https://auravcs.com/mobile - Skills: https://auravcs.com/skills - Stats (build in public — live GitHub stars, releases, project age): https://auravcs.com/stats - Docs: https://auravcs.com/docs - CLI: https://auravcs.com/cli - Machine-readable summary for agents: https://auravcs.com/AGENTS.md - Every URL: https://auravcs.com/sitemap.xml - Open source monorepo — app, CLI, engine, extension (GitHub): https://github.com/Naridon-Inc/aura - Download (free public beta): https://auravcs.com/#download - Privacy policy: https://auravcs.com/privacy ## Head-to-head comparisons (fair, with the other tool's strengths first) - AI editors and coding agents: https://auravcs.com/vs/cursor · /vs/claude-code · /vs/cline · /vs/kilo · /vs/devin · /vs/github-copilot-agent - Parallel agent orchestrators: https://auravcs.com/vs/conductor · /vs/superset · /vs/paseo · /vs/vibe-kanban · /vs/crystal · /vs/claude-squad - Terminal agents: https://auravcs.com/vs/aider · /vs/opencode · /vs/codex-cli · /vs/gemini-cli · /vs/amp · /vs/openhands - Editors, terminals and Git tooling: https://auravcs.com/vs/windsurf · /vs/zed · /vs/warp · /vs/git · /vs/graphite ## Open source alternatives (licence, platforms and pricing stated as the vendor states them) - Alternative to Conductor: https://auravcs.com/alternatives/conductor - Alternative to Superset: https://auravcs.com/alternatives/superset - Alternative to Cursor: https://auravcs.com/alternatives/cursor - Alternative to Windsurf: https://auravcs.com/alternatives/windsurf - Alternative to Devin: https://auravcs.com/alternatives/devin - Alternative to GitHub Copilot: https://auravcs.com/alternatives/github-copilot - Alternative to Graphite: https://auravcs.com/alternatives/graphite - Alternative to Warp: https://auravcs.com/alternatives/warp ## Guides — running several agents on one repository - Agent collision detection (stopping two agents editing the same function): https://auravcs.com/learn/agent-collision-detection - The agent inbox (how parallel agents talk to each other): https://auravcs.com/learn/agent-inbox - One Git worktree per agent: https://auravcs.com/learn/git-worktree-per-agent - AGENTS.md and CLAUDE.md, which instruction file does what: https://auravcs.com/learn/agents-md-vs-claude-md - How to review a pull request an AI wrote: https://auravcs.com/learn/review-ai-generated-pull-requests - Intent poisoning (when the commit message and the diff disagree): https://auravcs.com/learn/intent-poisoning - Handing context from one agent to the next: https://auravcs.com/learn/multi-agent-handoff-context ## Guides — reading code as structure - Semantic diff vs git diff: https://auravcs.com/learn/semantic-diff-vs-git-diff - Semantic diff for Rust (traits, impls, macros): https://auravcs.com/learn/rust-semantic-diff - Tree-sitter and the AST, explained: https://auravcs.com/learn/tree-sitter-ast-explained - AST merge vs text merge: https://auravcs.com/learn/ast-merge-vs-text-merge - Shadow branches (tool metadata outside your history): https://auravcs.com/learn/shadow-branch - A Merkle graph of your code: https://auravcs.com/learn/merkle-graph-code - Structural code search (patterns, not strings): https://auravcs.com/learn/structural-code-search ## Guides — governance, cost and self-hosting - An immutable audit log for AI-written code: https://auravcs.com/learn/immutable-audit-log - Running coding agents on an air-gapped network: https://auravcs.com/learn/air-gapped-ai-coding - Code provenance (tracing a line to the session that wrote it): https://auravcs.com/learn/ai-code-provenance - Stopping agents deleting code nobody agreed to delete: https://auravcs.com/learn/deletion-accountability - What an agent session actually costs, per response: https://auravcs.com/learn/ai-coding-cost-per-response - A self-hosted stack for AI-assisted development: https://auravcs.com/learn/self-hosted-ai-coding-stack ## Guides — agent loops & the autonomous work-loop - Aura Crew vs Claude Code loops (the autonomous work-loop, compared): https://auravcs.com/learn/aura-crew-vs-claude-code-loops - Autonomous coding agent loops explained (turn-based, goal, time, proactive): https://auravcs.com/learn/autonomous-coding-agent-loops-explained - How to run multiple coding agents in parallel without merge conflicts: https://auravcs.com/learn/run-multiple-coding-agents-in-parallel - Goal-based agent loops — giving a coding agent a definition of done: https://auravcs.com/learn/goal-based-agent-loops-definition-of-done - How to prove an AI code change did what you asked: https://auravcs.com/learn/prove-ai-code-changes - Agent work-loop harnesses compared (Aura Crew, Ralph, Beads, the bash loop): https://auravcs.com/learn/agent-loop-harnesses-compared ## Contact - Support / founder: mo@auravcs.com - Book a 30-minute call: https://cal.com/mo-naridon/30min ## Reference notes - Aura MCP (the 29-tool Model Context Protocol server): https://auravcs.com/learn/aura-mcp - Masking secrets in sync errors and diffs: https://auravcs.com/learn/mask-secrets-in-sync-errors-and-diffs - What belongs in an agent strategy inbox: https://auravcs.com/learn/agent-strategy-inbox-contents - dotenv duplicate keys (last one wins): https://auravcs.com/learn/dotenv-duplicate-keys-last-wins - Homebrew's default prefix on Apple silicon: https://auravcs.com/learn/homebrew-apple-silicon-default-prefix - tree-sitter grammar packages for TypeScript, TSX and Rust: https://auravcs.com/learn/tree-sitter-typescript-language-tsx-rust ## Status Public beta. Apache-2.0 across app, CLI, engine and extension in one public monorepo (https://github.com/Naridon-Inc/aura). macOS and Linux desktop builds, Windows in beta. Local by default: no account needed, and the collaboration server is self-hostable.