---
date: 2026-02-18
title: ⚙️ Config v1.3.4
version: "1.3.4"
---

- Added `create` field to `interface.prompts` and `interface.agents` configuration
  - Controls whether users can create new prompts/agents (default: `true`)
  - Complements existing `use`, `share`, and `public` permission fields
  - Boolean values still supported for backward compatibility
  - See [Interface Configuration](/docs/configuration/librechat_yaml/object_structure/interface#prompts) for details

- Added `interface.remoteAgents` configuration for remote agent access control
  - Added `use` setting to enable/disable remote agent usage (default: `false`)
  - Added `create` setting to control whether users can create remote agents (default: `false`)
  - Added `share` setting to control whether users can share remote agents (default: `false`)
  - Added `public` setting to enable public sharing of remote agents (default: `false`)
  - See [Interface Configuration](/docs/configuration/librechat_yaml/object_structure/interface#remoteagents) for details

- Added `models` and `inferenceProfiles` to `endpoints.bedrock` configuration
  - `models`: Specify custom model IDs available for Bedrock
  - `inferenceProfiles`: Map inference profile IDs to display names for cross-region inference
  - See [AWS Bedrock Configuration](/docs/configuration/librechat_yaml/object_structure/aws_bedrock#inferenceprofiles) for details

- Added Moonshot as a known endpoint
  - New `moonshot` known endpoint for Moonshot AI integration with icon support
  - See [Moonshot Configuration](/docs/configuration/librechat_yaml/ai_endpoints/moonshot) for details

- Added `deferred_tools` and `programmatic_tools` agent capabilities
  - `deferred_tools`: Enables lazy/event-driven loading of MCP tools, reducing context window usage (enabled by default)
  - `programmatic_tools`: Enables programmatic tool calling for MCP tools via code execution sandbox (disabled by default, requires latest Code Interpreter API)
  - See [Agents Configuration](/docs/configuration/librechat_yaml/object_structure/agents) for details

- Removed deprecated `forcePrompt` setting
  - Removed from Azure endpoint and custom endpoint configurations
  - Setting was deprecated and no longer functional

- Updated default model lists
  - Added `claude-sonnet-4-6` and `claude-opus-4-6` to Anthropic and Bedrock default models
  - Added `moonshotai.kimi` and `moonshotai.kimi-k2.5` to Bedrock default models
  - Added Bedrock parameter support for MoonshotAI, Z.AI, and OpenAI GPT-OSS models
  - Removed deprecated Claude models: claude-3-opus, claude-3-sonnet, claude-3-haiku, claude-2.x, claude-1.x, and claude-instant variants

- Added `artifacts` field to `modelSpecs` list entries
  - Enables the Artifacts capability for a model spec
  - Set to `true` to enable with default mode, or a mode string (e.g., `"default"`) for a specific mode
  - See [Model Specs Configuration](/docs/configuration/librechat_yaml/object_structure/model_specs#artifacts) for details

- Added `effort` parameter for Anthropic Adaptive Thinking in `modelSpecs` presets
  - Controls the thinking effort level for supported Anthropic models (e.g., Claude Opus 4.6)
  - Options: `"low"`, `"medium"`, `"high"`, `"max"`, or unset (model decides)
  - Supported by `anthropic` and `bedrock` (Anthropic models) endpoints
  - See [Model Specs Configuration](/docs/configuration/librechat_yaml/object_structure/model_specs#effort) for details