README

A Symfony AI platform bridge for the OpenCode Zen gateway.

Symfony ships one bridge package per provider — symfony/ai-open-router-platform, symfony/ai-deep-seek-platform and so on — but none for OpenCode. This package fills that gap so that OpenCode Zen can be used anywhere a Symfony\AI\Platform\PlatformInterface is accepted, including as the client behind the opencode-zen provider of mage-os/module-ai-base.

This is a plain PHP library. It contains no Magento code and needs no setup:upgrade.

Installation

composer require mage-os/library-ai-opencode-zen-platform

Usage

use MageOS\AiOpenCodeZenPlatform\Factory;

$platform = Factory::createPlatform($apiKey);

$result = $platform->invoke('kimi-k2.6', 'Why is the sky blue?');
echo $result->asText();

Point it somewhere else — a proxy in front of Zen, or a gateway run in-house that speaks the same wire format — with the baseUrl argument:

$platform = Factory::createPlatform($apiKey, baseUrl: 'https://ai.example.com/zen');

Get your API key from opencode.ai/auth.

Scope: Chat Completions only

Zen is a gateway, and it serves different model families from different endpoints. This bridge targets /v1/chat/completions only, which covers:

Family	Example ids
DeepSeek	`deepseek-v4-pro`, `deepseek-v4-flash`, `deepseek-v4.1-flash`
MiniMax	`minimax-m3`, `minimax-m2.7`
GLM	`glm-5.3`, `glm-5.3-flash`
Kimi	`kimi-k3`, `kimi-k2.7-code`
Free tier	`big-pickle`, `space-bunny-free`, `nemotron-3-ultra-free`, …

Model ids Zen serves from other endpoints cannot be reached through this bridge and will fail against the gateway:

Family	Zen endpoint
`gpt-`, `grok-`, `muse-*`	`/v1/responses`
`claude-`, `qwen`	`/v1/messages`
`gemini-*`	`/v1/models/<id>`
`jev-*`	`/v1/systemone`

For those, use symfony/ai-open-responses-platform, symfony/ai-anthropic-platform or symfony/ai-gemini-platform pointed at the matching Zen endpoint instead.

Embeddings are not wired: Zen exposes no /v1/embeddings endpoint.

Model catalogue

MageOS\AiOpenCodeZenPlatform\ModelCatalog accepts any model id and routes it to the chat-completions client. It deliberately enumerates nothing.

Every other bridge ships a static list of the models its provider served on release day. That works for a provider with a handful of long-lived names; Zen gains and deprecates models monthly and publishes its catalogue as an endpoint (https://opencode.ai/zen/v1/models, which is readable without a key). A frozen copy would reject a model an administrator can see in the gateway's own listing, so this bridge lets the gateway answer for its own catalogue.

The consequence is that an unreachable model id — gpt-5.5, say — is accepted locally and fails at Zen with Zen's own message, rather than being rejected here with a local one.

Versioning

symfony/ai-platform is experimental and makes no backward-compatibility promise. This package is verified against v0.13.0 and constrains itself to ^0.13; re-verify on upgrade.

mage-os / library-ai-opencode-zen-platform

Package info

Security

Statistics