FAQ

What is Agent Nine?

An autonomous AI agent for developers. Writes and edits code, runs commands, works with Git, connects to your systems (GitHub, Slack, Jira, etc.). Driven from your phone, runs on the desktop.

How is it different from Cursor / Claude Code / Copilot?

Who is it for?

• Developers (backend, frontend, mobile, DevOps) — any stack
• Freelancers and solo founders — accelerate solo work

How much does it cost?

3 paid tiers + Free. Roughly (messages per week):

• Free — ≈ 10 messages. No card, no expiry.
• Standard 1,990 ₽/mo — ≈ 500 messages.
• Pro 4,999 ₽/mo — ≈ 1,500 messages.
• Ultra 7,999 ₽/mo — ≈ 4,000 messages.

Full comparison — on the pricing page. Annual billing saves 25%.

Same features on every plan?

Yes. Every model (Sonnet, Opus, Haiku), every mode, every integration — available on any paid plan. The difference is usage volume and queue priority during peak hours.

How can I pay?

МИР, Visa, Mastercard cards, SBP, YooMoney, YooKassa. Legal entities — bank invoice.

Can I get a refund?

Yes, within 14 days of the first payment — no reason needed (152-ФЗ). Annual plans — pro-rated for unused months.

Can I change my plan?

Yes. Upgrade is instant, pro-rated. Downgrade applies from the next billing month. Cancel anytime — access stays until the end of the paid period.

Chat vs Code — what is the difference?

Chat — runs in the cloud, isolated sandbox with Python, Node.js, Go, Rust, Java. No access to local files. For architecture discussion, one-off scripts.

Code — runs through the desktop client with access to your files and terminal. The agent reads code, edits files, runs commands. For real development.

Integrations work in both modes.

Which models are available?

Three Claude models from Anthropic on every plan:

• Claude Sonnet 4.6 — default, 90% of tasks
• Claude Opus 4.7 — for complex architectural tasks (~6× more expensive than Sonnet by quota)
• Claude Haiku 4.5 — for simple quick questions (~5× cheaper than Sonnet)

Which programming languages does it work with?

Any. Agent Nine works with the filesystem and terminal, so the language is not a constraint. TypeScript, Python, Go, Rust, C++, Java, PHP, Ruby — anything you can run on the machine.

What are Plan mode and agents?

Plan mode — the agent first analyses the task and produces a step-by-step plan without changes, then executes (command /plan).

Agents — for large tasks Agent Nine spawns background agents that run in parallel. Three built-in types: General Purpose, Explore (read-only), Plan. Custom types can be defined in agentnine.md.

More — in the docs.

How does voice input work?

Microphone icon in the input field. Speak — Whisper transcribes Russian or English (auto-detect). Works in the web UI, desktop, and mobile browsers.

How do I install the desktop app?

Download from agentnine.io/download. Supported: Windows and Linux (x64/arm64). macOS and mobile apps are in development; for now use the web browser.

Does Chat mode require any install?

No. Chat runs in the browser at app.agentnine.io — no install, any device.

How do I sign in?

Sign in with a magic link: enter your email, get a one-time link (valid 15 minutes). No passwords.

Do I need a VPN?

No. Everything works directly from Russia — no foreign cards, no circumvention. The product also works from abroad.

What integrations are available?

11 built-in (OAuth): GitHub, Google Drive, Gmail, Google Calendar, Slack, Notion, Linear, Figma, Vercel, Sentry, Jira. Plus any custom MCP servers.

How do I connect them?

Settings → Integrations → “Connect” next to the service → OAuth flow. After that the agent has access to the integration’s tools.

Are my integration tokens secure?

OAuth tokens are stored encrypted (AES-256-GCM) on servers in Russia. Disconnect any integration with one click — the token is deleted and the agent loses access.

Can I connect my own MCP server?

Yes. Agent Nine implements the MCP protocol — connect any MCP-compatible server for your databases, APIs, custom tools. Instructions.

Is my data used to train models?

No. The Anthropic API runs zero-retention — your input and output are not stored and not used for training. Committed to in our Terms of Service.

Is my code safe?

In Chat mode code is sent only to the Anthropic API for processing; we store the dialogue history.

In Code mode files are read locally — only the needed context goes to the API — what the agent actually reads. Full codebases are not uploaded.

Agent Nine does not index your repos in the background.

Where is my chat history stored?

Chat history is stored on servers in Russia (Beget). Message text is sent to Anthropic only for the request; nothing is retained on their side.

Can I delete my account and data?

Yes. Settings → Account → Delete. Or write to support@agentnine.ru. Data is wiped within 30 days (plus 90 days in backups that overwrite automatically).

Can the agent delete my files?

Destructive commands (rm -rf, git reset --hard, DROP TABLE) are blocked automatically — Agent Nine asks for confirmation. You can undo every change of the current session with /undo.

How do I report vulnerabilities?

security@agentnine.ru — we reply within 48 hours.