“Can I clean here?”
Anatoly walks every file in your project, runs a 7-axis audit in parallel, and must prove each issue with grep, file reads, and a local semantic index before it ships to the report. One command. No code modified.
// one command, full audit, no code modified $ npx anatoly run → Scanner 247 files · TypeScript, Python → Triage 183 evaluate · 64 skip ($0) → RAG dual embed · code + nlp → Audit 7 axes · parallel · crash-isolated ● utility gemini flash ● duplication gemini flash ● correction claude sonnet ● overengineering gemini flash ● tests claude sonnet ● best practices claude sonnet ● documentation claude sonnet → Refinement 3 tiers · evidence required → Report sharded · severity-sorted ✓ 12 NEEDS_FIX ● 8 DEAD ● 4 DUPLICATE every finding has grep + file evidence attached
demo.gif available in repo · live preview soon
Linters catch syntax. They don't catch dead exports, hidden duplicates, over-engineered abstractions, or public APIs nobody documented. Manual review doesn't scale.
And no rule-based tool can honestly answer "is this function used anywhere?" — that question needs the whole project, not one file.
For every file, Anatoly runs seven audits in parallel — each routed to the model that fits it best. The agent must grep the codebase, read related files, and query the RAG index before any finding is reported. False positives are filtered through a 3-tier refinement pipeline.
Each axis is crash-isolated · routed to a different provider if you want · all 7 run in parallel
Full Anthropic agent runtime with tool use, real-time hooks (PostToolUse, Stop), and the Opus deliberation pass. The flagship integration.
Route axes to Gemini 2.5 Flash at zero token cost via Google OAuth. Reduces Claude calls by ~69% with circuit-breaker fallback.
Direct API access to virtually every modern provider — Anthropic, OpenAI, Google, Mistral, xAI, Groq, and more. Pick per axis.
Anatoly Cloud doesn't fork the product. The agent that runs scheduled audits in the cloud is the exact same one you run on your laptop.
The full product. No limits.
Run Anatoly on your machine through the CLI. Every feature is here: 7-axis audit, RAG semantic duplication, deliberation pass, auto-clean loop, watch mode, Claude Code hooks, dry-run estimator, sharded reports, Telegram notifications.
Same engine. GitHub-native. Always-on.
Anatoly Cloud connects to your GitHub repository and runs audits in the cloud — on schedule, on push, or on demand. No laptop needs to stay open. No per-developer setup. The team gets versioned reports tied to commits and PRs.
Cloud is runtime only. The engine is the same one you run locally.
Whether you run Anatoly locally or in the cloud, it runs with your API keys. The runtime is ours. The model choice and the spend stay with you.
Your Claude key. Your Gemini OAuth. Your OpenAI / Mistral / xAI keys. Your usage. Your bill.
Claude Code, Gemini CLI, or any API via Vercel AI SDK. Mix per axis.
Model calls go straight from the runtime to the provider. We never see the prompts or the output.
Better model lands tomorrow? Swap it in. Anatoly outlives any single vendor.
“A finding must be proven,
or it doesn't ship.”
Most AI code reviewers hallucinate. They flag a function as dead because they didn't trace the import. They flag duplicates because two names look alike.
Anatoly greps, reads, queries the RAG index, and cross-references the usage graph. Then a 3-tier refinement pass kicks in — auto-resolution, inter-axis contradiction detection, and an Opus investigator for the ambiguous cases. What you read in the report has been argued for.
If you have a question, an idea, a bug report, or you just want to see how a multi-LLM agent is built from the inside — come hang out.
No. Read-only audit. Auto-clean is opt-in and commits each fix individually so you can review and revert per-commit.
Platform fee only. You always bring your own model keys — local and cloud, same rule.
Yes. Always. AGPL-3.0. A commercial license is available for companies that cannot comply with AGPL.
Yes — exit codes (0 / 1 / 2), --plain mode for non-interactive pipelines, dry-run estimator, and SHA-cached re-runs at $0 on unchanged files.
TypeScript, Python, Rust, Go, Java, C#, Bash, SQL, YAML, JSON — auto-detected, with framework-aware evaluation for React and Next.js.
One email when Anatoly Cloud is ready. No drip campaign, no newsletter, no third-party lists. We use Resend (EU) and you can unsubscribe in one click.
“Can I clean here?”
$ npx anatoly run
Questions? The dev is on Discord.