Skip to content
View on GitHub

VS Code + Cline

Cline is an agentic coding extension for VS Code. It reads files, edits them, and runs commands autonomously. This is the validated setup for connecting Cline to Navigator.

Prerequisites

  • VS Code installed
  • The .env file created (see The .env Pattern) and the curl tests passing
  • The model ID you picked from the /v1/models response, and knowledge of whether it is a thinking-mode model
  • Your Navigator key handy

Installation

  1. Open VS Code.
  2. Open the Extensions panel (Ctrl+Shift+X on Windows/Linux, Cmd+Shift+X on macOS, or the four-square icon in the left sidebar).
  3. Search for Cline.
  4. Install the extension published in the VS Code left sidebar.
  5. A Cline icon appears in the VS Code left sidebar. Click it to open Cline.

Configuration

  1. With Cline open, click the gear icon in the top right of the Cline panel to open its settings.
  2. Set API Provider to OpenAI Compatible.

Why OpenAI Compatible and not LiteLLM. Cline offers a dedicated LiteLLM provider, but that provider tries to fetch model metadata from a LiteLLM admin endpoint that requires elevated permissions. UF IT-issued user keys don't have those permissions and get a 403 Forbidden error. OpenAI Compatible skips that metadata fetch and just calls the chat endpoint, which works fine.

Fill in the remaining fields:

Field Value
Base URL https://api.ai.it.ufl.edu/v1
API Key Your Navigator key (same value as ANTHROPIC_API_KEY in your .env)
Model ID The model ID you picked from the /v1/models response
Temperature Set to 1 IF your model is a thinking-mode model. Otherwise leave at Cline's default.

Thinking-mode models require temperature = 1. If your Navigator model runs in extended thinking mode, any request with temperature ≠ 1 is rejected with a BedrockException. Cline's default temperature is not 1, so you must set it explicitly. This is a Claude thinking-mode constraint, not a Cline or Navigator issue. If your model is NOT thinking-mode, leave temperature at Cline's default — no override needed.

  1. Save the configuration.
  2. Cline is now connected. Test with a simple prompt before running any real task.

Testing That It Works

Before pointing Cline at real AIBHS code, create a scratch directory and test in isolation. Cline is agentic and will read and modify files without much ceremony.

mkdir ~/cline-test && cd ~/cline-test
code .

In the new VS Code window, open Cline and enter a simple test prompt:

Create a file called hello.py that prints "hello from Cline" and then run it.

If Cline creates the file, asks permission to run it, runs it, and shows the output — chat, tool calls (write_file, execute_command), and streaming are all functional. You are done.

Known Caveats and Their Causes

  • Thinking-mode models are slow. Extended thinking models reason internally on every request. Not a Cline bug — expected behavior. If speed matters and your key is on a thinking-mode model, ask UF IT whether a non-thinking model is available.
  • Temperature setting is not optional on thinking-mode models. If it's missing, wrong, or reset to Cline's default after an extension update, every request fails. Verify temperature is 1 if a thinking-mode setup suddenly stops working.
  • Do not use the LiteLLM provider. It requires model-info admin permissions that Navigator user keys do not have. Use OpenAI Compatible instead.
  • Do not use the Anthropic provider. It hits api.anthropic.com directly, bypassing Navigator. Your Navigator key is not a valid Anthropic key and will fail.