Skip to main content

Bring Your Own Models (BYOM)

Connect and Manage Multiple AI Models within Pipefy and AI Agents.

Written by Natália Manffré

🔐 Available for all plans.

👤 For super admins, company admin and pipe admins.

🎯 Enables custom LLM provider keys integration to enhance AI Agents behaviors

What Is It and How to Use the BYOM Feature

⚠️ This feature is currently in research mode under phased rollout. It might not be available for your company yet, but it soon will.

Pipefy's BYOM (formerly BYOLLM) allows all organizations to connect multiple AI providers simultaneously and assign the specific LLM models for each Agent behavior.

This ensures:

  • Financial Efficiency: Use lightweight models for simple tasks and robust models for complex reasoning to optimize credit consumption.

  • Sovereignty & Security: Route sensitive data exclusively to your IT-approved providers.

  • Technical Differentiation: Plug in your own fine-tuned or proprietary models for specific business logic.

Note: It is important to highlight that this configuration applies only to AI Automations, AI Agents 2.0, and Assistant.

Before you start

  • You must be an Org Admin - Super Admin or Company Admin - to manage provider keys and the Model Registry at the Organization level.

    • Ensure you have your AI provider credentials (API keys, JSON, etc.) ready for the models you wish to homologate.

  • You must be a process builder - Pipe Admin - to select specific models within the Agent Builder.

List of supported LLM options

  • OpenAI:

    • gpt-4o, gpt-4o-mini, and dated release versions (e.g., gpt-4o-2024-08-06, gpt-4o-2024-11-20)

    • gpt-4.1-2025-04-14

    • gpt-5 and latest dated release versions (e.g., gpt-5-2025-08-07)

  • Azure OpenAI: Same models as OpenAI

  • Google Vertex:

    • gemini-1.5-pro, and gemini-1.5-pro-002

    • gemini-2.0-flash, and gemini-2.0-flash-lite

  • OCI Generative AI

    • xai.grok-4

    • meta.llama-4-scout-17b-16e-instruct

    • meta.llama-4-maverick-17b-128e-instruct-fp8

  • AWS Bedrock (check requirements below)

  • Custom Provider (check requirements below)

📌 The process of adding a new provider may vary, as specific fields and requirements depend on the chosen provider.

Requirements for AWS/Custom URL Providers

When using the “Custom” option (i.e., providing your own inference endpoint), ensure that your LLM and setup meet the following criteria:

  • OpenAI-Compatible Request and Response Format: Required so Pipefy AI Agents can correctly process the LLM’s output (e.g., choices, message, content).

  • Vision Support: Needed if you plan to use image and/or document input capabilities in your AI Agents.

  • Tool Calling Support: Ensures AI Agents can execute multi-step workflows and use agent skills effectively.

💡 Tips for using the feature:

  • Security: Keep API keys secure and avoid sharing them pubicly.

  • Limitations: Check the limitations and usage policies of the API you're integrating with to avoid service interruptions.

Step-by-Step: Setting up BYOM (Super Admin and Company Admin)

  1. Click on your Profile Picture and select Admin Panel.

  2. In the side menu, click on AI Settings.

  3. Click Bring Your Own Model to add a new provider configuration (e.g., OpenAI, Azure OpenAI).

    • Note: You can register multiple providers sharing the same model and vendor.

  4. Click on + Add New

  5. Enter a Name/Description for that model (e.g., "Legal Approved GPT-4").

  6. Input your provider credentials and click Save.

  7. Set Org. Default Model: Flag one model as the Default Model. This acts as the fallback for all tasks where a specific model is not selected.

    • Attention: Pipefy AI is the initial default and cannot be deleted, but it can be replaced as the default by your own model.

Step-by-Step: Assigning Models to Agent Behaviors (Pipe Admins)

  1. Open the Agent Builder (AI Agents) within your desired Pipe.

  2. Select or create a specific Behavior.

  3. Locate the Model dropdown at the very bottom.

  4. Choose the specific model you want to power this behavior (e.g., use a lightweight model for data extraction and a powerful model for decision-making).

  5. If you leave it as Organization Default, the Agent will use the model flagged by your Super Admin or Company Admin.

Limitations & Behaviors

  • Homologated Models: Currently supports existing homologated families (OpenAI, Azure OpenAI, Oracle OCI, Google Vertex) and custom models meeting specific requirements (OpenAI-compatible).

  • Mandatory Default: Each organization must always have exactly one default provider.

  • Model Deactivation: A model set as the default cannot be deactivated or deleted until another is set as the default.

  • Model Selection to Agent Behaviors only: Model selection applies exclusively to Agent Behaviors. The AI Automations and Assistant will run the set Org. Default Model, and Co-pilot always run on Pipefy AI for now.

  • Intelligent Document Processing (IDP): When using IDP as an advanced tool within an Agent, the system will continue to use Pipefy AI to process documents, even if a custom model is assigned to the behavior.

  • Web Search Tool: The Web Search advanced tool is also exclusively powered by Pipefy AI at this time, regardless of the custom model configuration.

Validation & Fallback Rules

Before publishing your Agent, verify the routing logic:

  1. Behavior Choice: The system first tries to use the specific model assigned to the behavior.

  2. Org Default: If none is assigned, it falls back to the Organization's default provider.

  3. Pipefy AI: If no custom default is found or if the configuration fails, it uses Pipefy AI as the final safety net.

Symptom

Cause

Fix

Model choice not appearing in Agent Builder

The model is registered but "Deactivated" in Admin Settings.

Go to AI Models and ensure the specific configuration is toggled to "Active".

Cannot delete a provider key

The provider is currently set as the "Organization Default".

Assign a different model as the Default first, then delete the old key.

Agent is using Pipefy AI instead of my key

Fallback triggered due to invalid credentials or model deactivation.

Check your API key status in the Model Registry and ensure the model is active.

🛠 Troubleshooting Common Issues:

  • Authentication Error: Verify that the API key is correct and active.

  • Inactive Integration: Make sure the third-party service is operational.

  • API Limits Exceeded: Check your quota, rate limits, and billing status directly in your AI provider's dashboard.

  • Model choice not appearing in Agent Builder: The model might be registered but "Deactivated". Go to AI Models and ensure the specific configuration is toggled to "Active".

  • Cannot delete a provider key: If the provider is currently set as the "Organization Default", assign a different model as the Default first, then delete the old key.

  • Agent is using Pipefy AI instead of custom key: A fallback was triggered due to invalid credentials or model deactivation. Check your API key status in the Model Registry and ensure the model is active.

If you have additional questions, contact Pipefy support or refer to the official documentation of the API you’re using.

For further details on pricing and plan limits, please visit the Pipefy pricing page

Did this answer your question?