OpenCode Config Flow: Complete Guide

This document explains how OpenCode configuration flows from user files through the plugin system to the Codex API.

Config Loading Order
Provider Options Flow
Model Selection & Persistence
Plugin Configuration
Examples
Best Practices

Config Loading Order

OpenCode loads and merges configuration from multiple sources in this order (last wins):

1. Global Config

~/.opencode/config.json
~/.opencode/opencode.json
~/.opencode/opencode.jsonc

2. Project Configs (traversed upward from cwd)

<project>/.opencode/opencode.json
<parent>/.opencode/opencode.json
... (up to worktree root)

3. Custom Config (via flags)

OPENCODE_CONFIG=/path/to/config.json opencode
# or
OPENCODE_CONFIG_CONTENT='{"model":"openai/gpt-5"}' opencode

4. Auth Configs

# From .well-known/opencode endpoints (for OAuth providers)
https://auth.example.com/.well-known/opencode

Source: tmp/opencode/packages/opencode/src/config/config.ts:26-51

Provider Options Flow

Options are merged at multiple stages before reaching the plugin:

Stage 1: Database Defaults

Models.dev provides baseline capabilities for each provider/model.

Stage 2: Environment Variables

export OPENAI_API_KEY="sk-..."

Stage 3: Custom Loaders

Plugins can inject options via the loader() function.

Stage 4: User Config (HIGHEST PRIORITY)

{
  "provider": {
    "openai": {
      "options": {
        "reasoningEffort": "medium",
        "textVerbosity": "low"
      }
    }
  }
}

Result: User config overrides everything else.

Source: tmp/opencode/packages/opencode/src/provider/provider.ts:236-339

Model Selection & Persistence

Display Names vs Internal IDs

Your Config (config/full-opencode.json):

{
  "provider": {
    "openai": {
      "models": {
        "GPT 5 Codex Medium (ChatGPT Subscription)": {
          "id": "gpt-5-codex",
          "options": {
            "reasoningEffort": "medium"
          }
        }
      }
    }
  }
}

What OpenCode Uses:

UI Display: “GPT 5 Codex Medium (ChatGPT Subscription)” ✅
Persistence: provider_id: "openai" + model_id: "gpt-5-codex" ✅
API Calls: "gpt-5-codex" ✅

TUI Persistence

The TUI stores recently used models in ~/.opencode/tui:

[[recently_used_models]]
provider_id = "openai"
model_id = "gpt-5-codex"
last_used = 2025-10-12T10:30:00Z

Key Point: Custom display names are UI-only. The underlying id field is what gets persisted and sent to APIs.

Source: tmp/opencode/packages/tui/internal/app/state.go:54-79

Plugin Configuration

How This Plugin Receives Config

Plugin Entry Point (index.ts:64-86):

async loader(getAuth: () => Promise<Auth>, provider: unknown) {
  const providerConfig = provider as {
    options?: Record<string, unknown>;
    models?: UserConfig["models"]
  };

  const userConfig: UserConfig = {
    global: providerConfig?.options || {},  // Global options
    models: providerConfig?.models || {},   // Per-model options
  };

  // ... use userConfig in custom fetch()
}

Config Structure

type UserConfig = {
  global: {
    // Applied to ALL models
    reasoningEffort?: "minimal" | "low" | "medium" | "high";
    textVerbosity?: "low" | "medium" | "high";
    include?: string[];
  };
  models: {
    [modelName: string]: {
      options?: {
        // Override global for specific model
        reasoningEffort?: "minimal" | "low" | "medium" | "high";
        textVerbosity?: "low" | "medium" | "high";
      };
    };
  };
};

Option Precedence

For a given model, options are merged:

Global options (provider.openai.options)
Model-specific options (provider.openai.models[modelName].options) ← WINS

Implementation: lib/request/request-transformer.ts:getModelConfig()

Examples

Example 1: Global Options Only

{
  "plugin": ["opencode-openai-codex-auth"],
  "provider": {
    "openai": {
      "options": {
        "reasoningEffort": "medium",
        "textVerbosity": "medium",
        "include": ["reasoning.encrypted_content"]
      }
    }
  }
}

Result: All OpenAI models use these options.

Example 2: Per-Model Override

{
  "plugin": ["opencode-openai-codex-auth"],
  "provider": {
    "openai": {
      "options": {
        "reasoningEffort": "medium",
        "textVerbosity": "medium"
      },
      "models": {
        "gpt-5-codex-high": {
          "name": "GPT 5 Codex High (OAuth)",
          "options": {
            "reasoningEffort": "high",
            "reasoningSummary": "detailed"
          }
        },
        "gpt-5-nano": {
          "name": "GPT 5 Nano (OAuth)",
          "options": {
            "reasoningEffort": "minimal",
            "textVerbosity": "low"
          }
        }
      }
    }
  }
}

Result:

gpt-5-codex-high uses reasoningEffort: "high" (overridden) + textVerbosity: "medium" (from global)
gpt-5-nano uses reasoningEffort: "minimal" + textVerbosity: "low" (both overridden)

Example 3: Full Configuration

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-openai-codex-auth"],
  "model": "openai/gpt-5-codex-medium",
  "provider": {
    "openai": {
      "options": {
        "reasoningEffort": "medium",
        "reasoningSummary": "auto",
        "textVerbosity": "medium",
        "include": ["reasoning.encrypted_content"]
      },
      "models": {
        "gpt-5-codex-low": {
          "name": "GPT 5 Codex Low (OAuth)",
          "options": {
            "reasoningEffort": "low"
          }
        },
        "gpt-5-codex-high": {
          "name": "GPT 5 Codex High (OAuth)",
          "options": {
            "reasoningEffort": "high",
            "reasoningSummary": "detailed"
          }
        }
      }
    }
  }
}

Best Practices

1. Use Per-Model Options for Variants

Instead of duplicating global options, override only what’s different:

❌ Bad:

{
  "models": {
    "gpt-5-low": {
      "id": "gpt-5",
      "options": {
        "reasoningEffort": "low",
        "textVerbosity": "low",
        "include": ["reasoning.encrypted_content"]
      }
    },
    "gpt-5-high": {
      "id": "gpt-5",
      "options": {
        "reasoningEffort": "high",
        "textVerbosity": "high",
        "include": ["reasoning.encrypted_content"]
      }
    }
  }
}

✅ Good:

{
  "options": {
    "include": ["reasoning.encrypted_content"]
  },
  "models": {
    "gpt-5-low": {
      "id": "gpt-5",
      "options": {
        "reasoningEffort": "low",
        "textVerbosity": "low"
      }
    },
    "gpt-5-high": {
      "id": "gpt-5",
      "options": {
        "reasoningEffort": "high",
        "textVerbosity": "high"
      }
    }
  }
}

2. Keep Display Names Meaningful

Custom model names help you remember what each variant does:

{
  "models": {
    "GPT 5 Codex - Fast & Cheap": {
      "id": "gpt-5-codex",
      "options": { "reasoningEffort": "low" }
    },
    "GPT 5 Codex - Balanced": {
      "id": "gpt-5-codex",
      "options": { "reasoningEffort": "medium" }
    },
    "GPT 5 Codex - Max Quality": {
      "id": "gpt-5-codex",
      "options": { "reasoningEffort": "high" }
    }
  }
}

3. Set Defaults at Global Level

Most common settings should be global:

{
  "options": {
    "reasoningEffort": "medium",
    "reasoningSummary": "auto",
    "textVerbosity": "medium",
    "include": ["reasoning.encrypted_content"]
  }
}

4. Use Config Files, Not Environment Variables

While you can set CODEX_MODE=0 to disable the bridge prompt, it’s better to document such settings in config files:

❌ Bad: CODEX_MODE=0 opencode

✅ Good: Create ~/.opencode/openai-codex-auth-config.json:

{
  "codexMode": false
}

Troubleshooting

Config Not Being Applied

Check config file syntax with jq . < config.json
Verify config file location (use absolute paths)
Check OpenCode logs for config load errors
Use OPENCODE_CONFIG_CONTENT to test minimal configs

Model Not Persisting

TUI remembers the id field, not the display name
Check ~/.opencode/tui for recently used models
Verify your config has the correct id field

Options Not Taking Effect

Model-specific options override global options
Plugin receives merged config from OpenCode
Add debug logging to verify what plugin receives

OpenCode Codex Auth Plugin

Access GPT-5 Codex through your ChatGPT Plus/Pro subscription