User guide / AI models
Install Playground GitHub ↗ Book Cloud

Self-host with the docs, star Enverif on GitHub, or book a Codefreex discovery call for managed cloud and custom plugins · MCP · skills.

Star on GitHub Book discovery call

AI models

Enverif uses a provider-neutral model contract and supports BYOK connections for OpenAI, Anthropic Claude, Google Gemini and DeepSeek.

As of 1.3.6, suggested model catalogs track current provider IDs (see changelog 1.3.3+). Capability-aware ModelRegistry feeds tools / vision / reasoning / structured-output flags into the provider catalog.

Explore samples in the live playground. Prefer managed cloud or a custom model gateway? Book a discovery call · Star on GitHub.

Each connection stores:

  • provider;
  • encrypted API key;
  • optional compatible base URL (for proxies, local models via OpenAI-compatible APIs);
  • default model identifier;
  • optional input/output prices per million tokens for run cost estimates.

The model field on an individual agent can override the connection default, allowing newly released model IDs to be used without an Enverif code change when the provider API remains compatible.

Supported providers

OpenAI

Suggested models include: gpt-5.4, gpt-5.2, gpt-5, gpt-5-mini, gpt-4.1, gpt-4.1-mini, gpt-4o, gpt-4o-mini, o3, o4-mini, o3-mini, o1, o1-mini

  • Reasoning models automatically receive reasoning_effort mapped from Enverif effort levels: Fast → low, Standard → medium, Deep → high.
  • Base URL: defaults to https://api.openai.com. Override for Azure OpenAI or compatible endpoints.
  • Supports images, PDFs, and text attachments.
  • Tool schema properties are normalized to JSON objects ({}) so providers do not reject empty PHP arrays.

Anthropic Claude

Suggested models include: claude-opus-5, claude-sonnet-5, claude-haiku-4-5, claude-opus-4-8, claude-sonnet-4-6, claude-opus-4-5, claude-sonnet-4-5

  • Uses the Anthropic Messages API (/v1/messages).
  • System prompt, vision and tool use all supported.
  • Fast / Standard / Deep maps to output_config.effort on supported generations.

Google Gemini

Suggested models track current Flash / Pro IDs (legacy 2.0 / 1.5 IDs are remapped).

  • Uses the Generative Language API (/v1beta/models/{model}:generateContent).
  • API key passed as URL query parameter.

DeepSeek

Suggested models: deepseek-v4-flash, deepseek-v4-pro

  • OpenAI-compatible /chat/completions endpoint at https://api.deepseek.com.
  • Retired aliases (deepseek-chat, deepseek-reasoner, …) remapped automatically.
  • Text-only attachments (no native image support).
  • V4 maps Enverif effort to reasoning_effort.

Custom model IDs

Select Custom model ID… in the model dropdown to enter any provider-compatible identifier, such as a fine-tuned variant or a newly released model not yet in Enverif's suggested list. Custom IDs bypass the suggested-model validation.

Using the connection in chat

In the chat composer, open Run settings to select Agent, Connection, Model and Effort for the current turn. Selecting Keep for this chat saves these as thread defaults so future turns reuse them automatically.

BYOK and pricing

Enverif never proxies your API keys through a third-party service. Keys are stored encrypted in your own database and sent directly from your server to the provider API. Input and output prices per million tokens are optional and used only for per-run cost estimates visible in the run console.

If credential decrypt fails after an APP_KEY change, re-enter the API key under AI Models — Enverif will ask for a fresh secret instead of crashing.