Fireworks AI

Put this in opencode.json in your project folder:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "fireworks": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Fireworks AI",
      "options": {
        "baseURL": "https://api.fireworks.ai/inference/v1",
        "apiKey": "{env:FIREWORKS_API_KEY}"
      },
      "models": {
        "accounts/fireworks/models/deepseek-v4-flash-0731": {
          "name": "accounts/fireworks/models/deepseek-v4-flash-0731"
        }
      }
    }
  }
}

• OpenCode reads any OpenAI-compatible address through the @ai-sdk/openai-compatible package, and an address that speaks the Responses API through @ai-sdk/openai.

From OpenCode documentation, read 2026-09-24.

Put this in ~/.pi/agent/models.json:

{
  "providers": {
    "fireworks": {
      "baseUrl": "https://api.fireworks.ai/inference/v1",
      "api": "openai-completions",
      "apiKey": "$FIREWORKS_API_KEY",
      "models": [
        {
          "id": "accounts/fireworks/models/deepseek-v4-flash-0731"
        }
      ]
    }
  }
}

• The apiKey field can name an environment variable as $NAME.

From Pi documentation, read 2026-09-24.

Put this in ~/.codex/config.toml:

model = "accounts/fireworks/models/deepseek-v4-flash-0731"
model_provider = "fireworks"

[model_providers.fireworks]
name = "Fireworks AI"
base_url = "https://api.fireworks.ai/inference/v1"
env_key = "FIREWORKS_API_KEY"

• Codex speaks the Responses API only: responses is the one supported wire API of a custom provider. Ollama and LM Studio are built in and start with --oss.

From Codex documentation, read 2026-09-24.

Put this in ~/.claude/settings.json, or variables in your shell:

export ANTHROPIC_BASE_URL="https://api.fireworks.ai/inference"
export ANTHROPIC_AUTH_TOKEN="$FIREWORKS_API_KEY"
export ANTHROPIC_MODEL="accounts/fireworks/models/deepseek-v4-flash-0731"
claude

• Claude Code sends Anthropic Messages requests to ANTHROPIC_BASE_URL. Anthropic says it does not support routing Claude Code to models other than Claude through any gateway, so some features may not work with another model.

From Claude Code documentation, read 2026-09-24.