Use CrofAI models (DeepSeek V4, GLM, Kimi, Qwen and more) natively in GitHub Copilot Chat with thinking effort control, tool calling, streaming reasoning and pricing.
Installation
Launch VS Code Quick Open (Ctrl+P), paste the following command, and press enter.
Use CrofAI models (DeepSeek V4, GLM, Kimi, Qwen, and more) natively in GitHub Copilot Chat.
Auto-discovery — every model from https://crof.ai/v1/models appears in the Copilot Chat model picker automatically. New models show up without updating the extension.
Thinking Effort — per-model None / Low / Medium / High control (default High) via Copilot Chat's native model picker configuration.
Streaming reasoning — thinking blocks stream inline as collapsible content.
Agent mode — full tool-calling support (file edits, terminal, search, MCP, skills).
Pricing — per-model cost ($ / 1M tokens) shown in the picker, refreshed from the live model list.
Secure key storage — API key stored in your OS keychain via VS Code SecretStorage.
Requirements
VS Code 1.134 or newer (uses Copilot Chat's native model-picker pricing support; 1.135+ shows the full pricing detail line). Copilot Chat is built into VS Code — no separate extension install needed.
Install this extension (or build from source — see below).
Run CrofAI: Set API Key from the Command Palette and paste your key. It is stored in your OS keychain.
Open Copilot Chat and pick a model from the CrofAI group.
Commands
Command
Description
CrofAI: Set API Key
Prompt for and store your CrofAI API key
CrofAI: Clear API Key
Remove the stored API key
CrofAI: Refresh Models
Re-fetch the model list from the API
CrofAI: Open Settings
Open this extension's settings
Settings
Setting
Default
Description
crofai-copilot.baseUrl
https://crof.ai/v1
API base URL (OpenAI-compatible)
crofai-copilot.maxTokens
0
Max output tokens per request (0 = no limit)
crofai-copilot.modelFilter
(empty)
Regex to filter auto-discovered models by id, e.g. ^glm-
crofai-copilot.defaultEffort
high
Default thinking effort when a request doesn't specify one
crofai-copilot.showPricing
true
Show per-model pricing in the model picker
crofai-copilot.debugMode
false
Log raw stream chunk fields to the output channel (for debugging empty/blank responses)
Building from source
npm install
npm run compile
Press F5 to launch the Extension Development Host. Package a .vsix with npm run package.
Notes
Tool-calling requests (agent mode, subagents) run with reasoning disabled — the CrofAI gateway returns an empty completion when reasoning_effort is combined with tools (confirmed with glm-5.3-flash). Plain chat keeps your chosen thinking effort.
Model discovery and pricing are fetched from https://crof.ai with your API key. If you point baseUrl at a self-hosted/proxied deployment, pricing may not be available — the extension hides it quietly in that case.
This extension relies on non-public Copilot Chat extension APIs (the same approach used by similar BYOK providers). It targets VS Code 1.134+; future VS Code versions may require updates.
