LEVEL11 in detail
One agent runtime underneath ten tools, three model providers, Korean integrations, and an account and billing platform.
Ten Open Tools
Pick a tool on the start screen. They share one workspace, one session and one agent, so a conversation started in the editor continues in the terminal tool.
The agent panel
Sessions, mode (Chat/Builder/Auto-Run), agent preset and model in one panel. Replies stream, tool calls fold into cards, thinking shows as a summary. The revert button beside each turn restores its checkpoint.
Agents and rules
Each agent has a name, a system prompt and a default model. Split roles — reviewer, test writer, docs — and switch per session. Rules merge global → user system prompt → project AGENT.md → .level11/rules.md.
Korean integrations
Services used in Korea come first, instead of Supabase or Stripe. Credentials are encrypted with the OS keychain (Electron safeStorage) and injected only into the commands the agent runs, as environment variables. They never appear in prompts, chat or files.
Three providers, one session
History is stored in a provider-neutral format. Start with Claude, continue with GPT, finish with Gemini — tool calls and results carry over. Claude uses adaptive thinking and effort, GPT reasoning effort, Gemini a thinking budget.
$ level11 · session 42
claude-opus-5-5 read_file src/index.ts … edit_file … run_command npm test
gpt-5.2 search_by_regex "TODO" … write_to_file docs/plan.md
gemini-3-pro web_fetch https://… … open_preview http://localhost:5173
// 한 세션, 세 제공자. 도구 호출과 결과는 그대로 이어집니다.Security
- In account mode, provider keys stay on the server. The app holds a device token; the server stores only its hash.
- BYOK keys and integration credentials are encrypted with safeStorage and stay on the device.
- Commands run after approval or from the allowlist. Auto-Run is enabled per session only.
- Every turn has a checkpoint, so changes can be reverted.
FAQ
Is an API key enough?
Yes. One key from OpenAI, Anthropic (Claude) or Google (Gemini) enables everything from model selection to agent runs. A LEVEL11 account is optional.
Which platforms?
macOS (Apple Silicon and Intel), Windows and Linux. Auto-update works on signed builds.
What about fonts?
The Korean code font defaults to D2Coding, bundled with the app. The English font defaults to a common code-editor font. Both can be chosen from the fonts installed on your machine.
Where is my data?
Sessions, settings and rules stay on your machine. In account mode the server records request metadata (tokens, cost) only, never the conversation.
Teams?
The Team plan has higher limits and included usage. Organisation seats and shared rules are in preparation.
Pricing
Start free, upgrade when you need to. All prices include VAT.
- Requests per day
- 3,000
- Included model usage
- $20/mo
- Models
- Fable · Opus · Sonnet · Haiku
- Requests per day
- 20,000
- Included model usage
- $80/mo
- Models
- Fable · Opus · Sonnet · Haiku
Model cost beyond the included amount is converted from USD at the month’s rate and added to the next invoice. Requests stop at the monthly cap (Free $3, Pro $200, Team $1,000).