Google Antigravity release notes: SDK 0.1.5 makes agent flows traceable

Google Antigravity has no newer public app release than 2.2.1, and the CLI track still tops out at 1.0.16. The missed signal sits in the SDK track: google-antigravity 0.1.5 has been on PyPI since June 25, 2026, and the diff from v0.1.4 shows a more useful SDK for teams building agent workflows with traces and better source handling.

Google Antigravity is Google's agentic development platform: an environment where a coding agent can read files, use tools, and help inside a project. An SDK is a code library that lets developers build their own agent workflows in Python. OpenTelemetry is an open standard for traces, metrics, and logs, so an agent run can leave a trail that people can inspect later.

Source: PyPI: google-antigravity 0.1.5

Source: GitHub compare v0.1.4...v0.1.5

Source: Google Antigravity changelog

Google Antigravity release notes: what SDK 0.1.5 changes

This is not a big IDE launch. It is a developer release that makes SDK-based agent flows easier to observe and review.

The useful pieces in the verified diff:

• A new google/antigravity/utils/otel.py file and examples/deep_dives/observability_otel.py example add OpenTelemetry tracing for agent runs and subagents.
• BuiltinTools.READ_URL_CONTENT is added as read_url_content, and the tool joins the read-only tool list. That gives SDK flows an explicit capability for fetching and reading URL content.
• The web tools example moves from web search alone to both search_web and URL fetching.
• pyproject.toml bumps the package from 0.1.4 to 0.1.5 and adds an otel extra for OpenTelemetry dependencies.

For Nordic automation teams, the point is practical. If you build custom agent flows with the Antigravity SDK, a correct terminal answer is not enough. You also want to see which source the agent read, which tools it used, and where a person approves the next step.

Source: GitHub compare v0.1.4...v0.1.5

Source: PyPI release history

How this differs from the previous SDK post

Hammer's previous Antigravity SDK post covered 0.1.4 and the clearer web search example. SDK 0.1.5 moves the story forward: from finding web sources to making the agent run traceable.

That matters if you use Antigravity as more than a chat surface. A coding agent is an AI agent that can work in code, run tools, and propose changes. Once that agent can reach external sources, MCP servers, or custom Python tools, the team needs a simple run receipt: input, tools, source, decision, human checkpoint.

Source: Google Antigravity SDK product page

Source: GitHub README

Human step: update first, then ask the agent to review

The installation line in Google's SDK page and README is still:

pip install google-antigravity

The README also shows SDK use with GEMINI_API_KEY or Application Default Credentials for Gemini Enterprise Agent Platform. Keep keys and project access out of chat: environment variables, Google authentication, secret managers, and scoped permissions are enough for most first passes.

Source: Google Antigravity SDK product page

Source: GitHub README, installation and authentication

Short example: use the new Antigravity feature

Paste this after the project is already on the right SDK version and the agent can see the repository you want reviewed:

Review this Antigravity SDK project for the 0.1.5 changes.
Focus on OpenTelemetry tracing, read_url_content and SEARCH_WEB usage, and whether the agent still starts in read-only mode.
Do not edit files. Return a run receipt: current observability gap, one URL-fetching test to add, and the approval point before enabling writes.

Good output should show:

• Which files the agent read to understand the SDK flow.
• Where OpenTelemetry tracing is missing or already present.
• How read_url_content and SEARCH_WEB stay tied to sources.
• Which human approval point is needed before write tools are enabled.

What we will watch next

SDK 0.1.5 is worth covering because it complements the latest CLI notes. CLI 1.0.15-1.0.16 made background work and logs easier to follow. SDK 0.1.5 makes the same question programmable: can we see what the agent did, with which sources and which tools?

For Hammer readers, this fits best as a Tool Forge step: start with one focused SDK routine, connect tracing and source fetching, and put the approval boundary where the agent moves from analysis to change.

Source: GitHub Antigravity CLI releases

Source: Google Antigravity changelog