OpenCode API key setup

OpenCode API Key Setup: Base URL, Provider & Models

OpenCode with a third-party API needs three things: an API key, a base_url, and a provider config.

Pay as you go for Claude, Kimi, GLM, and more: get an API key, set base_url and provider in opencode.json, then confirm with a test message.

OpenCode is installed: configure it Not installed yet: start with installation

Example opencode.jsonJSON

opencode.json

{
  "$schema": "https://opencode.ai/config.json",
  "model": "aicodewith-anthropic/claude-opus-5-5",
  "provider": {
    "aicodewith-anthropic": {
      "npm": "@ai-sdk/anthropic",
      "name": "AI Code With Anthropic",
      "options": {
        "baseURL": "https://api.aicodewith.ai",
        "apiKey": "{env:AICODEWITH_API_KEY}"
      },
      "models": {
        "claude-opus-5-5": {
          "name": "Claude Opus 5.5"
        }
      }
    }
  }
}

OpenCode not installed?

This page configures the OpenCode terminal CLI. If opencode --version already prints a version, skip ahead to the quick setup.

Open the official OpenCode download page Already installed: continue

Basic install command

curl -fsSL https://opencode.ai/install | bash

Or install globally with npm (this exact command is commonly searched)

npm install -g opencode-ai@latest

Installation succeeds when opencode --version prints a version number.

Continue with AI Code With Read the official installation docs

Secure quick setup

Pick a model, copy the config, save the file, and verify the connection in about 3 minutes.

1. Choose a model

The current catalog contains 46 models.

2. Copy the config

Provider: aicodewith-anthropic
npm adapter: @ai-sdk/anthropic
Base URL: https://api.aicodewith.ai
Model ID: claude-opus-5-5

{
  "$schema": "https://opencode.ai/config.json",
  "model": "aicodewith-anthropic/claude-opus-5-5",
  "provider": {
    "aicodewith-anthropic": {
      "npm": "@ai-sdk/anthropic",
      "name": "AI Code With Anthropic",
      "options": {
        "baseURL": "https://api.aicodewith.ai",
        "apiKey": "{env:AICODEWITH_API_KEY}"
      },
      "models": {
        "claude-opus-5-5": {
          "name": "Claude Opus 5.5"
        }
      }
    }
  }
}

Next: save the config

3. Save opencode.json and set the API key

Back up an existing opencode.json, then merge both model and provider from the generated content. Keep other Providers and store the real key only in a local environment variable.

Open the API key page

Save the config file

1. Open your system terminal
2. Run the command below to open the config file (created automatically if missing):
```
notepad "$env:USERPROFILE\.config\opencode\opencode.json"
```
3. Paste the config copied in step 2 and save it
Press Ctrl+S on Windows/Linux, Cmd+S on macOS.

Set the API key (replace your-key with your real key)

Temporary (only valid in the current terminal session)

$env:AICODEWITH_API_KEY = "your-key"

Lost when the terminal closes; good for a quick test.

Persistent (written to user config; new terminals pick it up)

setx AICODEWITH_API_KEY "your-key"

Open a new terminal or reload your shell config, then restart OpenCode.

4. Restart and verify

Run in your system terminalopencode
Type inside OpenCode/models confirm the Provider and Model ID match the page selection
Send in the chatReply with exactly: connection successful
confirm OpenCode displays a reply and match its time and model in AI Code With usage records

Usage records confirm that a request reached AI Code With. Determine success from the status, response content, and the reply actually shown in OpenCode.

Config did not load: troubleshoot

How does OpenCode connect to a third-party API?

OpenCode uses a custom Provider to send requests to the API service you choose: the Provider is the namespace that tells OpenCode which service to call, the API key handles authentication, the Base URL is the third-party API endpoint, and the Model ID selects the actual model.

Once saved, requests flow along OpenCode → Provider → Base URL → model API. After a /models check and a short test message, you can match the request against your usage records by time and model.

If setup fails, check these 4 things first

Each item links to a dedicated troubleshooting article with detailed steps.

401 Unauthorized
View detailed troubleshooting →
Usually the API key is invalid, mistyped, or not read correctly. Check that AICODEWITH_API_KEY is set without extra spaces and that OpenCode starts from the same terminal.
Model Not Found
View detailed troubleshooting →
The Model ID does not match the Provider config, or the channel does not support the model. Compare the generated Model ID and Provider pair character by character.
Base URL / 404
View detailed troubleshooting →
Usually the Base URL is wrong. Copy the value generated on this page and do not append /v1, /v1beta, or invent an endpoint path.
Config not applied / Provider missing
View detailed troubleshooting →
Check the config file path, that the JSON is valid, and that OpenCode reloaded the file. Reopen OpenCode after saving, then run /models.

Frequently asked questions

What should the Base URL be?

Copy the Base URL generated by the quick setup above. Do not append /v1 or /v1beta manually.

Where is the OpenCode config file?

This page generates ~/.config/opencode/opencode.json. Paths can vary by install method and version, so compare it with the current OpenCode docs.

How do I confirm the setup works?

Run /models to select the model, send a short message, then match the request by time and model in AI Code With usage records.

Which models are supported?

Claude, GPT, Gemini, DeepSeek, GLM, Kimi, and Grok, among others. Click "View all models" to browse the full catalog.

Need full installation or troubleshooting docs?

See the OpenCode setup docs for full installation guides, every config parameter, the complete model list, and troubleshooting.

Open the OpenCode setup docs