Google ships Interactions API in GA as Gemini default with background agents

• Google made the Interactions API generally available as the new default interface for Gemini, according to Google's launch post and Logan Kilpatrick's rollout thread.
• The GA surface folds models and agents into one stateful API, with managed agents, background execution, and broader tool support called out by Google and Phil Schmid's feature list.
• Phil Schmid's announcement adds two concrete agent pieces missing from the shorter corporate post: an Antigravity remote Linux sandbox and dedicated coding-agent skills.
• Migration is already being packaged as a promptable skill, and Phil Schmid's migration post says it can rewrite an existing app against the new schema automatically.
• Rollout is not limited to AI Studio, because Logan Kilpatrick's Vertex reply says Vertex will ship it too, while another Kilpatrick reply says it is available to both developers and enterprise customers.

You can jump from Google's blog post to the quickstart, then over to the docs and the coding-agent skill. Phil Schmid's launch thread is also where the weirdly specific details show up, including the remote Linux sandbox, background=True, and Gemini Omni listed as "soon" rather than day-one.

The biggest product change is not a new model. It is the control plane. Google's post says Interactions is now the primary interface for Gemini models and agents, while Kilpatrick's thread calls it Google's new default API going forward.

That changes the framing from single-request model calls to a stateful interaction object that can hold context across turns and agent runs. The quickstart attached in Phil Schmid's post shows the same client.interactions.create(...) entrypoint being used for both a plain model prompt and an agent invocation.

The GA pitch centers on longer-running work.

Features called out across the launch posts include:

• Managed Agents, per Google's launch post
• background=True for async, long-running interactions, per Phil Schmid's feature list
• Expanded tool support, per Google
• Multimodal tool use and combination, per Phil Schmid
• Multimodal generation, per Wes Roth's recap

Google is clearly standardizing around agent runs that outlive a single HTTP request. That is the part most teams will remember a month from now.

The more concrete capabilities live in the developer-facing thread and screenshot, not the shorter brand post.

According to Phil Schmid's announcement, the agent side includes:

• Antigravity Agent with an isolated remote Linux sandbox
• Image generation with Nano Banana
• Music generation with Lyria 3
• Video support through Gemini Omni, marked as coming soon

The attached code example in Phil Schmid's post also shows environment="remote" on an agent call, which makes the sandbox model feel less like marketing copy and more like a real execution surface.

Google is also shipping the transition path as tooling. Phil Schmid's post says a global skill install gives coding agents the right rules, SDK patterns, and current model versions for Interactions.

The two notable commands are simple enough to quote directly:

• npx skills add google-gemini/gemini-skills --skill gemini-interactions-api --global
• /gemini-interactions-api migrate my app to use the new interactions api

According to the same post, that migration prompt applies every API change from the migration guide automatically and works across Antigravity, Claude Code, Cursor, Codex, and Copilot. In a reply, Phil Schmid also said the docs will now show Interactions by default and that new features and models should land there first.

The rollout is broader than a standalone dev preview. In replies, Kilpatrick said Vertex will ship Interactions too, and another Kilpatrick reply said the API is available to both developers and enterprise customers.

That closes one obvious question left open by the main announcement: whether this was mainly an AI Studio developer surface or a platform-wide API shift. The answer, at least from the launch-thread replies, is platform-wide.