OpenCode
A terminal agent configured per project by a checked-in file. It reaches Devotel AI through the OpenAI-compatible provider, so the whole configuration is one JSON block.
Install
curl -fsSL https://opencode.ai/install | bashThe installer puts the binary in ~/.opencode/bin. Make sure that directory is on your PATH:
export PATH=$HOME/.opencode/bin:$PATHPersist it in ~/.zshrc or ~/.bashrc.
Configure the provider
Create opencode.json in your project folder:
{
"$schema": "https://opencode.ai/config.json",
"provider": {
"devotel": {
"npm": "@ai-sdk/openai-compatible",
"name": "Devotel AI",
"options": {
"baseURL": "https://api.devotel.com/v1",
"apiKey": "{env:DEVOTEL_API_KEY}"
},
"models": {
"kimi-k3": { "name": "Kimi K3" }
}
}
}
}The {env:DEVOTEL_API_KEY} placeholder means the file holds no secret and can be committed.
Export your key
export DEVOTEL_API_KEY=dvt-YOUR_KEYStart it and select the model
opencodePress /models, then choose Devotel AI → Kimi K3.
Selecting the model is per project, and OpenCode remembers it. If a session ever answers
suspiciously fast or in a different voice, press /models and check which provider is selected.
Using it
Describe what you want; OpenCode plans, then asks you to approve each tool call — file writes, shell commands, searches — before it runs. The conversation is stored per project, so you can leave and come back to the same thread.
Because it re-sends the accumulated context each step, a long session’s requests get progressively larger. That is normal for agents and is the main driver of token spend; see how agents use the API.
Troubleshooting
| Symptom | Cause and fix |
|---|---|
opencode: command not found | ~/.opencode/bin is not on PATH |
Devotel AI missing from /models | opencode.json is not in the folder you started from, or has a typo |
401 | DEVOTEL_API_KEY unset in the shell that launched opencode |
404 / model_not_found | The model key must be exactly kimi-k3 |