Zed
Zed supports custom OpenAI-compatible providers for the Zed Agent, Inline Assistant, commit messages, and thread summaries.
1. Set the token
Zed generates environment variable names from the provider ID. For provider ID
voxvey, set:
export VOXVEY_API_KEY="<access-token>"
Restart Zed after adding the environment variable. You can also enter the key in Zed's provider settings UI instead of using an environment variable.
2. Add the provider
Open Zed settings and add:
{
"language_models": {
"openai_compatible": {
"voxvey": {
"api_url": "https://api.voxvey.com/v1",
"available_models": [
{
"name": "deepseek/deepseek-v4-flash",
"display_name": "Voxvey DeepSeek V4 Flash",
"max_tokens": 1000000,
"max_output_tokens": 384000,
"capabilities": {
"tools": true,
"images": false,
"parallel_tool_calls": false,
"prompt_cache_key": false,
"chat_completions": true
}
},
{
"name": "openai/gpt-5.5",
"display_name": "Voxvey GPT-5.5",
"max_tokens": 400000,
"max_output_tokens": 64000,
"capabilities": {
"tools": true,
"images": false,
"parallel_tool_calls": false,
"prompt_cache_key": false,
"chat_completions": false
}
}
]
}
}
}
}
The capabilities.chat_completions: false setting tells Zed to use the
Responses endpoint for openai/gpt-5.5.
3. Make Voxvey the default model
Add a default model selection:
{
"agent": {
"default_model": {
"provider": "openai_compatible",
"model": "voxvey/deepseek/deepseek-v4-flash"
}
}
}
If your Zed build shows the model with a different provider label, select the Voxvey model from the Agent Panel model picker and save settings from there.
Verify
Open the Agent Panel and send:
Reply with the exact word ready.
If the model is missing, confirm the API key is available to the local Zed
process and that the provider ID is voxvey.