BedrockAI

BedrockAI brings Amazon Bedrock into Unreal Engine. Give every character a conversation of its own: a system prompt that says who it is, a memory of what the player said, and tools it may ask the game to use - give an item, open a door, start a quest. Replies can stream in piece by piece for typewriter dialogue, and the whole memory can be saved with your save game and restored later.

Create Conversation sets up one dialogue per NPC with its model, temperature, reply length and history limit. Send Message waits for the full reply, Send Message Stream delivers it as it is written. Register game actions with Add Tool and a JSON schema: when the model wants one, On Tool Use lists the calls, your Blueprint decides what really happens, and Submit Tool Results lets the character react in its own words. The history keeps only the most recent messages and never separates a tool call from its result.

Ask and Ask Stream answer one-off prompts such as item descriptions, quest text, names and hints. Generate Image creates a texture from a prompt with Amazon Nova Canvas or Titan Image Generator models and also returns the PNG bytes. Embed Text and Cosine Similarity add vector search over lore, quests and dialogue. Moderate Text checks player names and chat with an Amazon Bedrock guardrail, and the same guardrail can protect every conversation automatically. Invoke Model Raw reaches any Bedrock model through its native JSON API.

You choose the model: any Bedrock model or inference profile that supports the Converse API, per conversation or in the project settings. Players get credentials from an Amazon Cognito identity pool, as guests or signed in through Set Login Token; static keys and the default AWS credential chain are available for development, dedicated servers and tools. Requests run on background worker threads with automatic retries, and every result arrives back on the game thread.

The documentation explains how to enable model access, a least-privilege IAM policy for the identity pool role, and how to keep costs and abuse under control with guardrails, token limits and an optional backend.

BedrockAI is part of the Alpha XP AWS collection for Unreal Engine 4.27 and 5.8. It includes the AWS SDK for C++ prebuilt for Win64, Linux and Android and ships with styled HTML documentation.

Key features

• Conversations per NPC: model, system prompt, temperature, max tokens, tools and
• Send Message (Converse) and Send Message Stream (ConverseStream, UTF-8 safe pieces)
• Tool calls: Add Tool with a JSON input schema, On Tool Use, Submit Tool Results
• Memory limited to the most recent messages, tool calls always kept with their results
• Save and restore memory with Get History As JSON / Load History From JSON
• Ask and Ask Stream for one-shot prompts
• Generate Image into a transient texture plus PNG bytes (Nova Canvas / Titan Image
• Embed Text (Titan Text Embeddings) and Cosine Similarity for vector search
• Moderate Text and automatic guardrails on every conversation (Amazon Bedrock Guardrails)
• Invoke Model Raw for any model-native API
• Token usage and latency on every reply
• Any Bedrock model or inference profile with Converse support, chosen per conversation
• Amazon Cognito identity pools for guest and signed-in players; static keys and the
• Background worker threads, automatic retries, timeouts; cancellable latent nodes;
• C++ API with a native callback

Blueprint & C++ API

Every call is reachable from more than one place — pick the entry point that fits your project:

Callable runs when executed · Pure no exec pin, returns a value · LatentAsync finishes later and fires its outputs

BedrockAI 2

Conversation 6

Media 1

Config 3

Identity 5

Events & delegates

Bind these in Blueprint (drag off the object → Bind Event) or with AddDynamic in C++.

Project settings

Configure once under Project Settings › Plugins › BedrockAI — values are saved to the project's config .ini, so they travel with the project and your build.

Options & enums

Compatibility

One ready-made build per engine version is included in the download — copy the one that matches your project.

Technical details

What's in the download

Quick start

1. Unzip the download and pick the build that matches your engine (4.27 · 5.8).
2. Copy the BedrockAI folder into your project's Plugins/ folder (create it if needed).
3. Open the project, enable BedrockAI under Edit › Plugins and restart the editor.
4. Fill in Project Settings › Plugins › BedrockAI.
5. Right-click in any Blueprint graph and search BedrockAI — or include the public headers from C++.
6. Package for your target platform (Windows, Linux, Android).

Good to know