Integrations

Connect a chat app, your editor or a coding agent to tokspur, with direct API settings and local gateway instructions for Claude Code and Codex.

A terminal connects by cable to a local engine box through a partly inserted plug.

Client fit

Keep your chat app, editor or terminal and connect through a client that accepts a custom OpenAI-compatible endpoint. For a direct Chat Completions connection, use the base URL http://127.0.0.1:18081/v1, the model identifier tokspur and the key in .tokspur/state/api-key.

The clients below document these connection options. Streaming, tool calls and cancellation need to pass the connection checks with the exact client, model and runtime versions.

Choose your tools

App, editor or agentConnection
Chat appsOpenAI-compatible provider
VS CodeCustom Endpoint, Cline or Continue
CursorCline extension
JetBrains IDEsAI Assistant or Continue
ZedOpenAI-compatible provider
Roo Code and Kilo CodeOpenAI-compatible provider
Neovim and EmacsCodeCompanion or gptel
AiderDirect Chat Completions
OpenCodeDirect Chat Completions
Claude CodeMessages through the local gateway
Codex CLIResponses through the local gateway

The guides follow each client’s documented provider settings. Use the connection checks to confirm the route with your client version. Tokspur serves text and tool calls. These connections do not add image input, embeddings or inline completion support.

Load the key

Run this from the tokspur checkout in the shell that will launch your client. It loads the configured key without printing it:

export TOKSPUR_API_KEY="$(cat .tokspur/state/api-key)"

For a graphical editor, use its provider key field or credential storage. Keep keys out of shared configuration files. The direct API base is http://127.0.0.1:18081/v1.

Chat apps

A chat app that lets you add an OpenAI-compatible provider can use tokspur as a private assistant. Add a provider with the base URL http://127.0.0.1:18081/v1, the model tokspur and the key from .tokspur/state/api-key. Set the context length to 65536 and turn off image and file uploads, because tokspur works with text only.

Run the app on the same Mac, because the server listens on the loopback interface only. tokspur answers requests in turn, so it suits a personal assistant rather than a shared service. It refuses requests that carry a browser Origin header, so use a desktop or terminal app, or a web app whose own server makes the requests.

Cursor

For a local connection inside Cursor, use the Cline extension and configure its OpenAI Compatible provider. Cline lists Cursor among its supported editors. This uses Cline as the agent inside Cursor.

Cursor's built-in model requests pass through Cursor's servers, including requests made with your own API key. That is a different privacy boundary from a direct local connection. See Cursor's API-key documentation before choosing that route.

See Cline setup

Cline

In Cline settings, select OpenAI Compatible as the API provider, set the base URL to http://127.0.0.1:18081/v1, paste the key and set the model ID to tokspur. Set the context window to 65536, reserve up to 8192 output tokens and turn off image input. Cline documents custom API connections, including local endpoints.

Review tool permissions and network settings before using an agent on a private repository.

Continue

In the Continue extension for VS Code or JetBrains, add the model to your existing config.yaml with the OpenAI provider, the local apiBase and the key. Use chat and edit roles and leave the Responses API off. Continue documents custom local endpoints and selecting Chat Completions instead of the Responses API.

models:
  - name: tokspur
    provider: openai
    model: tokspur
    apiBase: http://127.0.0.1:18081/v1
    apiKey: YOUR_KEY
    useResponsesApi: false
    roles: [chat, edit]
    capabilities: [tool_use]
    defaultCompletionOptions:
      contextLength: 65536
      maxTokens: 8192

Aider

After loading the key, Aider reads the base URL and key from the environment and takes the model with openai/ in front of its identifier. Follow its custom API instructions, and pick an edit format that matches the model.

export OPENAI_API_BASE=http://127.0.0.1:18081/v1
export OPENAI_API_KEY="$TOKSPUR_API_KEY"
aider --model openai/tokspur

Connection checks

Before enabling an agent
CheckExpected behaviour
Text requestThe model tokspur returns a response through the local endpoint.
StreamingThe client assembles chunks in order, keeps thinking apart from the answer and recognises completion.
Tool round-tripThe client receives a tool call, returns the result with the matching call identifier and keeps declaring the tools used earlier in the conversation.
CancellationStopping a request releases the generation before another begins.
ErrorsA wrong key (401), a refused request (400) and a busy server (429) are visible to the user.

Agent permissions

The client decides which files to read, commands to run and external services to contact. Review those permissions in the client. Choosing a local model endpoint does not alter them.

Read the API contract before configuring a client. Use connection troubleshooting when a request fails.