GoogleAntigravity

LiteRT model configurations, dynamic GCP environment hydration, and performance updates

Thursday, 9 July 2026

AI summary

Written by AI from the official notes. Check them for exact details.

This release enhances local execution capabilities and multimodal tool outputs for Gemma models.

Expanded local execution with LiteRT for Gemma models.
Custom tools can now return multimodal media directly.
Integrated built-in web fetch tool for structured content.
Modernized MCP string prefix to resolve naming issues.
Improved compatibility with Python 3.14.

Why it matters: Developers using Gemma models should care for improved local execution and multimodal capabilities.

Official notes on antigravity.google

Full release notes8 changes

This release expands local execution capabilities by broadening support for multimodal and web-connected tool workflows. Developers can now run Gemma models locally using LiteRT or integrate with local OpenAI-compatible APIs, natively return multimodal media from custom tools, and leverage a more robust, streamlined lifecycle hooks framework that is fully aligned with Model Context Protocol (MCP) workflows.

Local Model Connectivity: Introduced LiteRTAgentConfig and LiteRTConnectionStrategy for LiteRT-LM (supporting local Gemma execution), LocalOpenAIAgentConfig and LocalOpenAIConnectionStrategy for OpenAI-compatible APIs (supporting Ollama and LM Studio), and a background loopback HTTP translation server.
Multimodal Tool Outputs: Enabled custom tools to return media assets (Image, Document, Audio, Video) directly via a single tool response without needing separate follow-up turns (supplemental_media).
Built-in Web Fetch Tool: Integrated the read_url_content tool end-to-end for fetching structured web content natively with the ReadUrlContentResult Pydantic model.
MCP String Prefix Modernization: Decoupled tool calls and safety policy engines from legacy "mcp_" string synthesis, resolving name mismatch issues by utilizing explicit server_name attributes in tool evaluation.
Python 3.14 Compatibility: Resolved namespace class conflicts and typing normalization issues in agent.py and public_api_test.py under Python 3.14 deferred annotations evaluation.
Vertex Validation Errors: Cleared a misleading reference to API keys in VertexEndpoint validation error messages, limiting fields to project and location.
OTel Trace Warnings: Resolved detached contextvars warnings and set-status race conditions by removing use_span context managers from Turn/Session hooks and checking span recording readiness.
Exception Wrapping Mapping: Fixed agent_middleware integration check failures by making the example check for error message substrings instead of strict exception types.

Full release notes8 changes

Local Model Connectivity: Introduced LiteRTAgentConfig and LiteRTConnectionStrategy for LiteRT-LM (supporting local Gemma execution), LocalOpenAIAgentConfig and LocalOpenAIConnectionStrategy for OpenAI-compatible APIs (supporting Ollama and LM Studio), and a background loopback HTTP translation server.
Multimodal Tool Outputs: Enabled custom tools to return media assets (Image, Document, Audio, Video) directly via a single tool response without needing separate follow-up turns (supplemental_media).
Built-in Web Fetch Tool: Integrated the read_url_content tool end-to-end for fetching structured web content natively with the ReadUrlContentResult Pydantic model.
MCP String Prefix Modernization: Decoupled tool calls and safety policy engines from legacy "mcp_" string synthesis, resolving name mismatch issues by utilizing explicit server_name attributes in tool evaluation.
Python 3.14 Compatibility: Resolved namespace class conflicts and typing normalization issues in agent.py and public_api_test.py under Python 3.14 deferred annotations evaluation.
Vertex Validation Errors: Cleared a misleading reference to API keys in VertexEndpoint validation error messages, limiting fields to project and location.
OTel Trace Warnings: Resolved detached contextvars warnings and set-status race conditions by removing use_span context managers from Turn/Session hooks and checking span recording readiness.
Exception Wrapping Mapping: Fixed agent_middleware integration check failures by making the example check for error message substrings instead of strict exception types.