Connect with a setup command
OpenCode, Claude Code, Codex, and Cline support a terminal command with browser approval. You need the selected tool and Node.js 22 or newer installed, plus owner or admin access to a Limitless workspace. For Cline, use Node.js 22 or 24. Cursor and Cline’s manual fallback use the guided settings flow below, which does not require Node.js or a terminal command. The automatic flows are enabled per deployment. The setup page shows which connections are available, including Cline’s manual fallback when its command is disabled. Live model checks used Claude Code 2.1.226 and Codex CLI 0.162.1 on macOS; Windows setup has not been qualified.- Choose your tool in the console and copy its setup command.
- Run it in your terminal. A browser window opens to Limitless; sign in if prompted.
- Match the connection code in your browser with the one in your terminal.
- Choose the workspace and project that should pay for requests, then select Approve connection.
- Return to the terminal and wait for setup to finish. Follow its instructions to start a coding session.
OpenCode
OpenCode uses its native provider login:Claude Code
When enabled, run:node ~/.config/limitless/claude-code/start.mjs to refresh the compatible model picker before Claude Code starts. Plain claude uses the last configured list. Model picker configuration requires Claude Code 2.1.242 or newer.
Setup configures your global Claude Code user settings for Limitless, retaining unrelated settings and making a backup of the previous file. It respects CLAUDE_CONFIG_DIR when set; otherwise it uses ~/.claude/settings.json. A local credential helper supplies the API key. The key is kept separately from the settings file and is not placed in the setup command or browser URL. Project or managed settings can override user settings.
The model is GLM 5.3 Flash Unrestricted, including requests made through Claude Code’s small-model setting. This connection does not provide a Claude model.
Codex CLI
When enabled, run:CODEX_HOME and preserves your existing default configuration. Start a session with:
multi_agent = false). Text, function tools, and Codex’s custom apply_patch tool are supported. Setup targets Codex CLI; desktop setup is not included in this flow.
Cline extension
Install the Cline extension in your editor. When automatic setup is enabled, run the command shown on the connection page:/v1/models source; review a newly discovered model’s configuration against the connection page before use.
- Wait for setup to finish, then reload the editor.
- In Cline, choose the Limitless provider and your chosen model. On a fresh installation’s welcome screen, first choose Bring my own API key, then Limitless.
- If Plan and Act use separate models, choose Limitless in both modes.
- Start a small task and check its request in Limitless Activity.
@cline/core 0.0.93 on macOS. The existing isolated extension retained its Limitless selection and completed a live read/edit/test task against the hosted model. All four requests settled with no outstanding holds; the temporary connection key was revoked and its stored credentials cleared after testing. A separate earlier fresh-install test verified choosing Limitless and automatically filling the key, endpoint, model and limits. These checks qualify the browser-approval and SDK setup flow on the tested macOS installation.
Manual settings for Cursor or Cline
Choose Cursor, or choose Cline → Set up manually instead, on the connection page. If Cline’s command is disabled, its manual settings appear directly. Select the workspace and project in the console, then create a dedicated key or use an existing project key. New keys expire after 90 days. Copy a newly created key before leaving the page; it cannot be displayed again later. Copy the base URL and model ID into your tool’s settings as described below. These guided flows use the tools’ supported settings interfaces. They do not import settings automatically. On October 16, Cursor 3.12.30 accepted the saved key, endpoint and selected custom model, but its Free plan blocked the test prompt before inference. A paid-plan live editor qualification remains outstanding. Cline’s automatic setup and live editor task passed as described above; manually pasting every provider field was not part of those tests.Cursor
Before creating a key, sign in to Cursor and check that your plan supports custom API keys. Cursor Free does not support this setup, even when a key is saved and the custom model appears selected. Our October 16 test was blocked with “Named models unavailable. Free plans can only use Auto.” Use a plan that allows custom models; your organization must also allow OpenAI API keys.- Open Cursor Settings → Models → API Keys.
- Paste your Limitless key into OpenAI API Key, save it, and enable Use OpenAI API Key.
- Enable Override OpenAI Base URL and set it to
https://trylimitless.ai/v1. - Add and enable a custom model with the exact ID
deepseek-v4.1-flash-unrestricted. - Restart Cursor, open a new chat, and select that custom model.
- On the Limitless connection page, confirm that you selected the model, then choose Open a test chat in Cursor. Check the model picker and send the prefilled prompt. Look for the request in Limitless Activity.
Cline manual settings
- Open the Cline extension → Settings and choose OpenAI Compatible as the API Provider.
- Paste your Limitless key into the API-key field.
- Set Base URL to
https://trylimitless.ai/v1and choose a current Model ID on the connection page, includingdeepseek-v4.1-flash-unrestricted. Copy that model’s context/output limits and image-support setting. - Open Model Configuration and set the values below explicitly.
- If you use separate models for Plan and Act, select and configure this model in both modes. Save the settings and start a new task.
Cline’s custom-model defaults can differ from these limits and may enable images. The table shows DeepSeek Flash settings. GLM 5.3 is text-only with a 1,000,000-token context; use the current connection page values. See Cline’s OpenAI Compatible guide. A separate proxy, Cline CLI installation, or Node.js installation is not needed for these settings.
Kilo
Kilo remains marked Coming soon until its connection is qualified. Use the connection page for current availability.Model discovery and switching
All setup paths use the same available-model catalog as the API, filtered by the protocol each tool uses. Newly registered models need no separate model-ID update in each installer. Models disabled in the deployment or awaiting required pricing are omitted.
The current models support Chat Completions, Messages and Responses. The connection page supplies the models and limits appropriate to each coding tool.
The connection page shows each model’s current context/output limits and whether it accepts images. Coding-tool output budgets are capped at 32,768 tokens or the model’s smaller limit. See API compatibility.
Keys and usage
Each new key created by a connection flow is named after the tool. It expires after 90 days and uses the selected workspace’s credits, spending caps, and request limits. It has normal project inference permissions; it is not limited to a single model. Reusing an existing key retains that key’s current expiry and permissions. Manage or revoke connection keys in API keys. To replace a revoked or expired key, repeat your tool’s connection flow and update its saved key. Terminal connection codes expire after ten minutes. If automatic setup is interrupted after browser approval, check API keys and revoke any unused key before retrying. Creating a key does not confirm that an IDE is configured. Send a request from your tool and check Activity for its outcome. Connecting does not start or reserve model capacity. The selected model must be available to accept requests.Other OpenAI-compatible tools
For tools with a custom OpenAI-compatible provider setting, create a project API key and use:
Check that your tool supports Chat Completions, streaming, and function tools. A custom base URL alone does not guarantee compatibility with every API feature. See Compatibility.