MelisAIEngineGemini

Google Gemini provider for the Melis AI engine — package melisplatform/melis-ai-engine-gemini.

Purpose

MelisAIEngineGemini plugs Google Gemini into Melis AI. It is a provider implementation: it extends MelisAIEngine's abstract MelisAIEngineModelService and supplies the Gemini-specific HTTP client, request payload (the contents + parts schema), tool/function calling (functionDeclarations), file handling (File API or inline embed) and token-usage parsing. It ships no UI of its own — agents, instances and scenarios are configured in the AI backoffice, and the engine routes a request here whenever the active model's company is Google. See Melis AI and MCP & tools for the surrounding workflow.

Enable it

This is a standard Laminas module (module name MelisAIEngineGemini). Install it with Composer (composer require melisplatform/melis-ai-engine-gemini) and add it to config/melis.module.load.php:

'MelisAIEngineGemini',

It requires melisplatform/melis-ai-engine (^5.3) and PHP ^8.1|^8.3. Its dbdeploy deltas register the Google company and the Gemini models. At runtime, file handling can also use MelisDocumentUpload (when the model's "internal upload" option is on). The provider is only used if the module is in the active modules list — the engine checks this before instantiating it.

How the engine selects it

MelisAIEngineService::getActiveModelClass($company, $modelId, $agentId) (in melis-ai-engine) inspects the active model's macp_company_name. When it contains Google and MelisAIEngineGemini is active, it builds MelisAIEngineModelGeminiService via the service manager (build() with modelId / agentId options). No manual wiring is needed.

Key services

Service alias	Role
MelisAIEngineModelGeminiService	The Gemini provider. Subclass of MelisAIEngine\Service\MelisAIEngineModelService. Built per request with modelId + agentId.

Notable methods of MelisAIEngineModelGeminiService:

Method	Role
setClient()	Builds the Laminas HTTP client; URI = {api_url}/models/{mam_generative_model}:generateContent?key={mapk_keys}, POST, 3600s timeout.
sendCustomAI(?array $payload = [])	Sends the payload; parses candidates/parts; dispatches functionCalls (MCP or MelisAIEngineFunctionService); retries on 5xx/429; returns result, errors, responseData, payload, needs_continuation, session_id, tool_results.
continueConversation($sessionId, $continuationContext, $existingContext = [])	Rehydrates contents from the PHP session (or a fallback context) and re-calls sendCustomAI().
addToolsToPayload($payload)	Adds tools[].functionDeclarations + toolConfig.functionCallingConfig (mode AUTO); when no functions, adds the url_context tool instead.
addContentToPayload($payload, $role, $prompt, $files, $content) / constructContent($role, $prompt, $files)	Build a contents entry (role + parts: text, fileData, inlineData).
processFiles($files) / processContextFiles($files)	Normalise uploads per the model's upload mode (fileapi / embed) and the internal-upload flag.
setPromptTokenCount() / setResponseTokenCount() / setTotalTokenCount()	Read usageMetadata.{promptTokenCount,candidatesTokenCount,totalTokenCount}.
getAllowedMimetypes() / getMessageKey()	Return the configured allowed MIME types / the message key (contents).

Constants: FILEAPI_UPLOAD_MODE = "fileapi", EMBED_UPLOAD_MODE = "embed".

File upload modes

The active model's mam_file_upload_mode selects how files reach Gemini:

• fileapi — uploads to the Gemini File API (upload_url), then references the returned fileUri. The service waits until the file state is ACTIVE.
• embed — sends file bytes inline as base64 (inlineData), with text extraction for .docx and .xlsx.

When mam_internal_upload is on, text files are routed through MelisDocumentUpload instead (so they can be re-displayed in past chat sessions), and only non-text files go to the File API / embed.

Configuration

Gemini endpoints and limits live under the config key plugins → melisaiengine → datas → AI → Gemini (see config/app.interface.php):

Key	Default
api_url	https://generativelanguage.googleapis.com/v1beta
upload_url	https://generativelanguage.googleapis.com/upload/v1beta/files
allowed_mimetypes	image / pdf / text / office / video / audio / zip MIME list

The API key is not stored in config — it comes from the model row (mapk_keys), managed in the AI backoffice.

Backoffice

None of its own. The module declares the melisaienginegemini plugin shell (asset bundle config) but ships no tool, controller action, view helper or dashboard plugin. All AI screens are provided by melis-ai / melis-ai-engine.

Database tables

The module owns no tables; it populates melis-ai-engine's tables via dbdeploy:

Table	What this module inserts
melis_ai_companies	The Google company row.
melis_ai_models	Gemini models (e.g. gemini-2.5-flash, gemini-2.5-pro, gemini-2.5-flash-lite, gemini-3.1-pro-preview) with their mam_generative_model, upload mode and file flags.

Key files

Concern	Path
Provider service	vendor/melisplatform/melis-ai-engine-gemini/src/Service/MelisAIEngineModelGeminiService.php
Service factory	vendor/melisplatform/melis-ai-engine-gemini/src/Service/Factory/MelisAIEngineModelGeminiServiceFactory.php
Service alias / routing	vendor/melisplatform/melis-ai-engine-gemini/config/module.config.php
Endpoints & allowed MIME types	vendor/melisplatform/melis-ai-engine-gemini/config/app.interface.php
dbdeploy (company + models)	vendor/melisplatform/melis-ai-engine-gemini/install/dbdeploy/
Module bootstrap	vendor/melisplatform/melis-ai-engine-gemini/src/Module.php

See also

• Melis AI — agents, instances, scenarios.
• MCP & tools — function/tool calling.
• Module reference — the full module map.