Skip to main content
This page covers common issues when connecting 4096bytes to clients. Start by checking three values: API Key, Base URL, and model name.

Connection parameters

OpenAI compatible clients usually use https://api.4096bytes.com. If the dashboard shows a client-specific address in the Use key dialog, use the value shown there.
Do not manually append API paths. Most clients only need the Base URL root, and the client app adds the request path. Claude Code, Codex, and Cursor may use different formats, so follow the matching client page.
Copy the model name from the dashboard. Keep letters, numbers, hyphens, and version suffixes exactly the same. Do not copy model names from old screenshots or other users.
YOUR_4096BYTES_API_KEY and PASTE_MODEL_NAME_FROM_DASHBOARD are placeholders. Copy the real API Key and model name from the dashboard. The fixed https://api.4096bytes.com value is the current OpenAI compatible endpoint.

API Key management

Technically yes, but it is not recommended. Use separate keys for Codex, Claude Code, Cursor, and Cherry Studio. This makes usage easier to trace and lets you disable only one key if it leaks.
Usually no. Copy and store it immediately after creation. If you forgot to save it, delete the old key and create a new one.
Delete the leaked key in the dashboard immediately. Then create a new key and update the affected client. Do not keep using a key that appeared in screenshots, logs, chats, or public repositories.
For personal long-term use, you can start without a limit. For temporary tests, team members, shared devices, or rarely used clients, set a small limit to avoid unexpected usage.

Errors

The API Key may be invalid, deleted, incomplete, or copied with extra characters. Copy it again and check for spaces, newlines, smart quotes, or hidden characters.
Check the key group, model availability, and account status. The group must be enabled and include the model you are calling.
The Base URL may be wrong, or the client may be appending paths incorrectly. Use https://api.4096bytes.com and remove any manually added full request path.
Request rate, concurrency, balance, or key limits may have been reached. Reduce concurrency, retry later, and check balance, quota, and rate limits in the dashboard.
Copy the model name again from the dashboard. It must match exactly, including capitalization, numbers, hyphens, and suffixes.
The Base URL format is incomplete. It must start with https://. Do not enter api.4096bytes.com, extra spaces, quotes, or line breaks.

Client behavior

Restart the client first. Then check whether there are multiple config locations, providers, or model lists. Confirm the active model comes from 4096bytes.
Cursor custom API Keys may not affect every built-in feature. Select the 4096bytes model in Chat or Agent. Tab completion, indexing, and background tasks may still use Cursor services.
Claude Code reads Anthropic-related environment variables. Follow Claude Code and use ANTHROPIC_BASE_URL plus the matching auth variable.
Put the new config at the beginning of ~/.codex/config.toml, and make sure model_provider points to the 4096bytes provider. Restart the terminal after saving.
If the client cannot fetch the model list automatically, add the model name manually from the dashboard and select it again in the current chat.
  1. Copy the API Key again.
  2. Confirm the Base URL is https://api.4096bytes.com or the client-specific value shown in the dashboard.
  3. Copy the model name again.
  4. Confirm the provider type matches the client page.
  5. Restart the client or terminal.
  6. Check balance, quota, key expiry, rate limits, and group status.
  7. Contact support with redacted details.
If you have not completed the basic setup, start with Get an API Key, then open the matching client page.