Use FantasyAI as a native language-model provider inside VS Code's Copilot Chat.
The extension connects to https://fantasyai.cloud/api/v1 and exposes its bundled FantasyAI chat catalog in the Copilot model picker — with streaming, tool calling, and vision where supported. Requires VS Code 1.116 or later and Copilot Chat.
Features
Hardcoded FantasyAI endpoint — no setup beyond your API key
Native Copilot integration — models appear in Copilot's Language Models picker
Instant model catalog — models and their capabilities are bundled with the extension
New models — GPT 5.6 Sol, Terra and Astra, Claude Opus 5, Claude Sonnet 5, and Qwen 3.8 Flash
Native thinking — reasoning streams into Copilot's collapsible thinking block
Secure key storage — API key kept in VS Code's encrypted SecretStorage
Streaming responses — real-time token streaming with stop button
Quick start
Install the extension (.vsix or from the marketplace).
Open the Command Palette (Ctrl+Shift+P) → FantasyAI: Set API Key.
Paste your FantasyAI API key. The gateway connection is checked automatically.
Pick a model from the Copilot model picker, or run FantasyAI: Pick Model.
That's it — no endpoint or provider configuration required.
Commands
Command
Description
FantasyAI: Set API Key
Save or update your FantasyAI API key
FantasyAI: Clear API Key
Remove the stored key
FantasyAI: Pick Model
Set the active FantasyAI model
FantasyAI: Refresh Models
Reload the bundled model list and check the gateway connection
FantasyAI: Test Connection
Verify the API is reachable
Settings
Setting
Default
Description
fantasyAI.activeModel
""
Currently active model
fantasyAI.temperature
0.2
Sampling temperature (0–2)
fantasyAI.maxTokens
8192
Maximum output tokens per response
fantasyAI.systemPrompt
"You are a helpful…"
System prompt
fantasyAI.requestTimeoutMs
120000
Request timeout (ms)
fantasyAI.showReasoning
true
Show reasoning in Copilot's native thinking block
The API key is not stored in settings.json — it lives in VS Code's encrypted SecretStorage.
Native thinking uses LanguageModelThinkingPart, following DeepSeek V4 for Copilot. This API is still proposed; when a host does not expose it, reasoning is hidden while answers and tool calls continue to work. Hiding reasoning also preserves it for tool continuations.
Model names and routing follow FantasyAI's catalog. Its GPT 5.6 Sol/Terra/Astra and Claude Opus 5 entries are FantasyAI display identities routed through DeepSeek V4 by the gateway.
