Integrations
Toss Payments, NICEPAY, PayPle, Cafe24, NHN Cloud, NAVER Cloud, Megazone, Vercel, and Anthropic/OpenAI keys for project code. Credentials are encrypted with the OS keychain and passed only as environment variables.
What can be connected
The Integrations tool puts the services used in Korea first. Enter credentials for a service and the commands the agent runs can use them as environment variables.
| Category | Service | Environment variables |
|---|---|---|
| Payments | Toss Payments | TOSS_CLIENT_KEY, TOSS_SECRET_KEY |
| NICEPAY | NICEPAY_CLIENT_ID, NICEPAY_SECRET_KEY |
|
| PayPle | PAYPLE_CST_ID, PAYPLE_CUST_KEY, PAYPLE_AUTH_KEY (optional) |
|
| Hosting | Cafe24 | CAFE24_MALL_ID, CAFE24_CLIENT_ID, CAFE24_CLIENT_SECRET, CAFE24_SSH_HOST (optional) |
| NHN Cloud | NHN_APP_KEY, NHN_SECRET_KEY, NHN_REGION |
|
| Cloud | NAVER Cloud | NCP_ACCESS_KEY, NCP_SECRET_KEY, NCP_REGION |
| Megazone Cloud | MZC_PROVIDER, MZC_ACCESS_KEY, MZC_SECRET_KEY, MZC_REGION |
|
| Deploy | Vercel | VERCEL_TOKEN |
| AI | Anthropic, OpenAI | ANTHROPIC_API_KEY, OPENAI_API_KEY |
The AI entries are keys for your project's code: the app the agent is building calls Claude or GPT with them. They are separate from the key LEVEL11 itself uses to call models (the BYOK key under Settings → LEVEL11 account).
How credentials are handled
- Values are encrypted with the OS keychain (Electron safeStorage: Keychain on macOS, DPAPI on Windows) and stored on this device only. They are never uploaded to LEVEL11's servers.
- They are injected as environment variables only into commands the agent runs with
run_command. They are not written into prompts, chat history or files. - The agent's system prompt lists which services are connected and which variable names are set, never the values, and tells the agent not to echo, log or write them to files.
- In the settings screen a stored secret shows only as
•••• (unchanged); entering a new value replaces it. - Disconnect deletes the stored credentials from this device.
Integrations are currently used by the agent only. Other parts of the IDE (the editor, terminal tabs) do not read them. Commands you type yourself in the Terminal tool do not get them either, so either let the agent run the command or set the variables yourself.
Working with the agent
A connected service reaches the agent as one line, "the environment variables for this service are set for run_command", plus a link to the official docs. So a request like "write an example that calls the Toss payment-confirmation API and run it" produces code that reads TOSS_SECRET_KEY from the environment instead of embedding it, and the agent runs the command to check the result. The 커머스 빌더 (Commerce Builder) preset carries these conventions in its system prompt: secrets from the environment, payment confirmation, cancellation and webhook verification on the server side, and a clear line between test and live keys. Presets are explained in Rules and MCP.
Start with test keys (test_ck_…, test_sk_…), confirm the flow, then switch to live keys. On a device holding live keys, watch what Auto-Run executes more closely: dangerous commands still ask for approval there, but a payment API call is an ordinary command.