Editor setup

Choose your editor

Start tokspur with the setup guide, then use the connection below. The API base is http://127.0.0.1:18081/v1, the model is tokspur and the key is in .tokspur/state/api-key. Use text and tool calls, with a total context of 65,536 tokens including output. Inline completion, images and embeddings need separate model support.

• VS Code
• JetBrains
• Zed
• Roo Code
• Kilo Code
• Neovim
• Emacs
• Other editors

VS Code

Run Chat: Manage Language Models, choose Add Models, then Custom Endpoint. Select Chat Completions and enter the tokspur key through the key prompt. In the generated chatLanguageModels.json, configure the model as below, retaining the generated secret reference.

[
  {
    "name": "tokspur",
    "vendor": "customendpoint",
    "apiKey": "${input:tokspurApiKey}",
    "apiType": "chat-completions",
    "models": [
      {
        "id": "tokspur",
        "name": "tokspur",
        "url": "http://127.0.0.1:18081/v1/chat/completions",
        "toolCalling": true,
        "vision": false,
        "maxInputTokens": 57344,
        "maxOutputTokens": 8192
      }
    ]
  }
]

Select tokspur in Chat. If your VS Code build or organisation does not offer Custom Endpoint, use Cline or Continue. Copilot inline completions use separate settings.

VS Code model settings

JetBrains

In IntelliJ IDEA, PyCharm, WebStorm, Rider or another supported JetBrains IDE, install AI Assistant and open Settings, Tools, AI Assistant, Providers & API keys. Add an OpenAI-compatible provider with the tokspur base URL and key, enable tool calling and use Test Connection.

Select tokspur in AI Chat. Assign it manually under Models Assignment for the supported editor features, setting the context to 65536. Agent availability depends on the IDE and agent. AI completion is configured separately and expects a completion model. Continue is another route through its JetBrains extension.

JetBrains custom model setup

Zed

Open agent: open settings and add an OpenAI-compatible provider named tokspur. Enter the base URL, model identifier and key in the provider UI. The equivalent model settings are:

{
  "language_models": {
    "openai_compatible": {
      "tokspur": {
        "api_url": "http://127.0.0.1:18081/v1",
        "available_models": [{
          "name": "tokspur",
          "display_name": "tokspur",
          "max_tokens": 65536,
          "max_output_tokens": 8192,
          "capabilities": {
            "tools": true,
            "images": false,
            "parallel_tool_calls": false,
            "chat_completions": true,
            "max_tokens_parameter": true
          }
        }]
      }
    }
  }
}

Keep the key in Zed’s provider settings, then select tokspur in the Agent Panel. Edit prediction uses separate settings. External agents keep their own provider configuration.

Zed provider setup

Roo Code

In the Roo Code settings panel, select OpenAI Compatible. Enter the tokspur base URL, key and model identifier. Set the context window to 65536 and maximum output to 8192. Turn off image support and computer use.

Roo uses native tool calling. Complete a tool request and return its result before using it for a longer task.

Roo Code provider setup

Kilo Code

Open Settings, Providers, Custom provider. Use tokspur as the provider ID and display name, choose OpenAI Compatible, and enter the base URL and key. Fetch the models or enter tokspur manually, save and select it in the model picker. Use a 65536-token context and an 8192-token output limit where available.

For Kilo Desktop, add a Custom AI Provider under AI settings with the same Chat Completions connection. Leave image support off.

Kilo provider setup

Neovim

With CodeCompanion installed, extend its openai_compatible adapter. Export TOKSPUR_API_KEY in the shell that launches Neovim, using the key setup. This adapter takes the server URL without /v1.

require("codecompanion").setup({
  adapters = {
    http = {
      tokspur = function()
        return require("codecompanion.adapters").extend("openai_compatible", {
          formatted_name = "tokspur",
          env = {
            url = "http://127.0.0.1:18081",
            api_key = "TOKSPUR_API_KEY",
          },
          schema = { model = { default = "tokspur" } },
        })
      end,
    },
  },
  interactions = {
    chat = { adapter = "tokspur" },
    inline = { adapter = "tokspur" },
  },
})

Open :CodeCompanionChat and send a short request. Keep the conversation within the server’s total context limit.

CodeCompanion adapter setup

Emacs

With gptel installed, load the following configuration. Start Emacs from a shell with TOKSPUR_API_KEY set, or use gptel’s supported credential storage.

(setq gptel-model 'tokspur
      gptel-backend
      (gptel-make-openai "tokspur"
        :host "127.0.0.1:18081"
        :protocol "http"
        :endpoint "/v1/chat/completions"
        :key (lambda () (getenv "TOKSPUR_API_KEY"))
        :stream t
        :models '(tokspur)))

Run M-x gptel to open a conversation. Tool use requires gptel tool configuration and the connection checks below.

gptel configuration

Other editors

Cursor can use Cline. VSCodium can use an available Open VSX build of Cline or Continue. For Xcode, Sublime Text, Vim and other editors, run Aider or OpenCode in a terminal in the same project directory. That route does not change the editor’s built-in assistant.

If an extension asks for an OpenAI-compatible URL, choose Chat Completions and use the shared settings. A client that runs in a remote container or cloud cannot reach the Mac through its own loopback address. Run the client locally alongside tokspur.

Finish with the connection checks, including cancellation and a tool round-trip.