Antigravity now has built-in browser.. finallyy :) And you can attach any file format now too.. One very annoying issue - The agent doesn't seem to be aware that their is an inbuilt browser and always starts the preview or any other browser work in Chrome by default.. i hope that get fixed ! i saw it today and apparently it had shipped on 27th.. good to see them updating agy multiple times a week now..

# Antigravity Features and Practical Usage 🚀 "If you say it works, show me." Antigravity's agent opens Chrome itself, clicks through your app, and records video proof that the feature works. 📌 Title and Feature URL Title: Browser URL: 📝 Overview Browser is Antigravity's integrated Chrome capability. The agent actually opens a browser, drives your web app, and saves a browser recording as an Artifact. Instead of a text claim, you get video evidence that the implementation truly works. 🔧 How It Works - The agent opens Chrome and autonomously clicks through flows in your web app - The session is captured as a browser recording, serving as video proof the feature works - Recordings are one kind of Antigravity Artifact, alongside rich markdown, diagrams, and code diffs - Using the specialized Browser Subagent, the agent launches an app on localhost, clicks the UI, fills forms, and verifies the real user experience - On errors it reads the console logs, debugs the cause, and retries, all without human intervention - Combined with web-oriented MCP servers, it can also do autonomous research through the browser and DevTools-level inspection 🛠 Practical Usage - Hand E2E testing to the agent and receive a browser recording Artifact as evidence - Automatically attach screenshots and recordings to review after an implementation - Add a web MCP for autonomous cross-page information gathering - Offload tedious manual UI checks like form submissions and page transitions 🎯 Use Cases - Verify a new feature's E2E flow with a recording and attach video proof to the PR - Delegate regression checks for critical flows like login and checkout - Let the agent self-debug and retry starting from a console error - Use a web MCP to research reference material and spec pages across sites ⚠️ Caveats - Browser automation requires Google Chrome, and the agent runs inside a completely separate Chrome profile from your normal one - Because it drives a real browser, handle target URLs and credentials with care - Autonomous control is convenient but risks unintended actions if pointed at production - Recording Artifacts consume storage per session, so define a management policy for CI integration #Antigravity# #DevTools#

# Antigravity Features and Practical Usage 🚀 Say "audit my dependencies every morning" once, and the agent keeps working in the background on its own. That is the promise of Scheduled Tasks in Antigravity 2.0. 📌 Title and Feature URL Title: Scheduled Tasks URL: 📝 Overview Scheduled Tasks let you define crons that trigger Antigravity agent invocations on a predefined schedule. You send timed instructions to your agents, have them run autonomously in the background, and get the results reported back later. The strength is that routine work can run without a human watching over it the whole time. 🔧 How It Works - The /schedule command supports both a run at a specified future time and recurring runs on a cron schedule - Schedules are given as a standard 5-field cron expression, and the remaining arguments are the prompt to run - Recurring example: `/schedule "0 9 * * 1-5" Run git fetch, prune stale local branches, and summarize pending PR reviews.` runs routine work at 9am on weekdays - For an hourly cadence you express the interval as a cron expression such as `"0 * * * *"` - The prompt you provide fires as an agent invocation at the scheduled moment, and results come back when the work completes - Introduced in Antigravity 2.0, it pairs with strengthened subagents and agent management to support autonomous, parallel work 🛠 Practical Usage - Register a task by following /schedule with a cron expression and then the instruction to run - Use a near-term cron expression for quick checks and a recurring one for routine work - There is no run-count cap, so keep cron intervals conservative to avoid needless executions - Queue batch-style investigations for off-hours when developers are away 🎯 Use Cases - Automate a morning audit of dependency vulnerabilities and updates - Analyze CI failure logs overnight to pre-diagnose root causes - Check a few minutes later whether a build or deploy has finished - Run a health check script on a cadence to catch issues early ⚠️ Caveats - Schedules assume a standard 5-field cron expression; natural language timing is not documented - There is no argument for capping the number of runs, so a poorly chosen interval can pile up needless executions - Scheduled runs still let the agent perform real actions, so pair destructive tasks with guardrails like Hooks - Details and limits may change by version, so confirm with the latest docs #Antigravity# #DevTools#

# Antigravity Features and Practical Usage 🚀 What if you could inject your own logic into an agent's execution loop? Antigravity's lifecycle hooks let you enforce policy, log everything, and add guardrails automatically. 📌 Title and Feature URL Title: Hooks URL: 📝 Overview Hooks let you run custom scripts or shell commands at specific points in Antigravity's agent execution loop. By reacting to events such as before/after a tool call or at session termination, you can enforce rules, capture diagnostics, and add guardrails without relying on a human to watch every step. They are the foundation for making agents follow your organization's policies. 🔧 How It Works Hooks fire on lifecycle events and differ by their nature: - PreToolUse: fires before a tool runs, letting you inspect arguments and allow or deny the call - PostToolUse: fires after a tool completes, useful for logging and audit - PreInvocation: fires before the model is called - PostInvocation: fires immediately after each model invocation completes - Stop: fires when the execution loop ends - By behavior, hooks fall into Inspect (read-only, non-blocking), Decide (read-only, blocking allow/deny), and Transform (can modify data) - Decide hooks return a decision result (HookResult) with allow/deny to stop policy violations before they happen 🛠 Practical Usage - Use PreToolUse to inspect risky command arguments and return deny to halt execution - Use PostToolUse to emit structured logs and build metrics or audit trails - Inject system instructions before a model call to apply internal guidelines to the prompt - Use Stop as a stop-condition check to force tests or lint to pass - Execution order is Decide -> operation -> Inspect, validating first to avoid TOCTOU-style gaps 🎯 Use Cases - Block destructive operations against a production DB before the agent runs them - Auto-generate an audit log by recording every tool call for later review - Inject coding standards and prohibited actions before each model call - Force lint, type checks, and tests to run at session termination ⚠️ Caveats - Inspect and Decide hooks are read-only and cannot mutate data (use Transform hooks for that) - Faulty deny logic can stop legitimate work, so design conditions carefully - The config location and JSON I/O contract may vary by version or distribution, so confirm with the latest docs - Hooks interrupt the loop, so heavy work inside them slows overall responsiveness #Antigravity# #DevTools#

# Antigravity Features and Practical Usage 🚀 A large refactor handled by several agents working at once. Subagents turn one agent into a team. 📌 Title and Feature URL Title: Subagents URL: 📝 Overview Subagents enable parallelized workflows. When the main agent receives a complex task, it automatically breaks it into subtasks and spawns independent subagents that execute concurrently. The finished results are then aggregated back into the conversation. 🔧 How It Works Subagent orchestration runs largely autonomously: - Environment profiling: it checks permissions, reads directories, and verifies dependencies. - Autonomous subagent definition: the orchestrator decides how to decompose the work without user configuration. - Parallel spawning: once dependencies are met, agents with isolated context windows run simultaneously. - Output synthesis: the orchestrator assembles each subagent's output into the final deliverable. - It also combines with background automation such as scheduled tasks. 🛠 Practical Usage - Hand over a large task and let the main agent split it into subtasks automatically. - Process work in parallel by package or module so independent changes proceed at once. - Let the orchestrator manage ordering, since subagents launch as dependencies resolve. - Have the finished diffs aggregated into the conversation for a single review. 🎯 Use Cases - Split a large refactor by package, run it in parallel, and aggregate the completed diffs into the conversation. - Advance several independent features at the same time to cut waiting time. - Run a bug bash that fans out across a broad surface to surface issues quickly. - Assign research and code changes to separate subagents that work in parallel. ⚠️ Caveats - The orchestrator decides decomposition autonomously, so review the results to ensure the split matches your intent. - Parallel execution involves multiple concurrent environments and model calls, so watch cost and rate limits. - Tasks with tight dependencies may see limited benefit from parallelization. #Antigravity# #Agents#

# Antigravity Features and Practical Usage 🚀 A "USB-C port for AI" that safely connects your agent to your databases and internal tools—that is Antigravity's MCP. 📌 Title and Feature URL Title: MCP URL: 📝 Overview Antigravity supports the Model Context Protocol (MCP), a standard that lets the editor securely connect to your local tools, databases, and external services. MCP acts as a "USB-C port for AI," standardizing how AI agents and large language models plug into different data sources. 🔧 How It Works MCP works as a standardized plug-in mechanism: - MCP is an open standard that lets AI agents securely connect to external tools and data sources. - Antigravity includes an MCP Store where you can search for and install the service you need. - Connection details (such as a Project ID) are entered through a simple form, and credentials—passwords or IAM credentials—are stored securely. - For example, the BigQuery remote MCP server enables running queries, getting metadata, and listing resources. - BigQuery's MCP uses OAuth 2.0 with IAM for authentication and authorization, supporting Google Cloud identities. 🛠 Practical Usage - Search the MCP Store for the service you need, such as "BigQuery," and click Install to begin setup. - Enter connection details like your Project ID in the form and register credentials securely. - Once connected, ask in natural language to inspect a table's schema, then have the agent generate a script from the result. - Mix and match local-process and remote-host MCP configurations as needed. 🎯 Use Cases - Install BigQuery from the MCP Store, inspect a billing table's schema, and generate a FinOps script. - Connect to an internal database and build queries from natural-language questions. - Have the agent understand your data platform's structure via metadata and resource listings. - Build automation workflows that span local tools and external services. ⚠️ Caveats - MCP grants the agent access to external data, so least-privilege IAM scoping is a prerequisite. - Credentials are stored securely from the form, but review the permission scope of each service you connect carefully. - Service-specific auth (OAuth 2.0 / IAM) and billing terms, as with BigQuery, should be confirmed before adoption. #Antigravity# #MCP#

# Antigravity Features and Practical Usage 🚀 Stop re-explaining your usual review steps to the agent every time. Skills package that knowledge into one reusable file. 📌 Title and Feature URL Title: Skills URL: 📝 Overview Skills are reusable packages of knowledge that extend what your agent can do. A central SKILL.md file tells the agent what the skill is, when to use it, and how to execute it—so it accomplishes tasks consistently without overwhelming its context window. 🔧 How It Works The heart of a skill is a single markdown file, SKILL.md: - SKILL.md is the "brain" of the skill, describing what it is, when to use it, and how to run it. - SKILL.md is the only required file, but you can bundle additional scripts, examples, and resource directories. - Skills placed in the global scope (~/.gemini/antigravity/skills/) are available across every project on the machine. - It suits general utilities like "Format JSON," "Generate UUIDs," or "Review Code Style." - Writing a small markdown file is enough to turn a common prompt into a reusable, team-shareable command. 🛠 Practical Usage - Create a code-review skill so the agent checks for bugs, style issues, and best practices when reviewing PRs. - Capture your team's review procedure in SKILL.md and place it under ~/.gemini/antigravity/skills/ to standardize it. - Bundle scripts and examples as resources to make the steps concrete. - Keep project-specific conventions in the workspace scope and general utilities in the global scope. 🎯 Use Cases - Standardize your team's code-review bar as a code-review skill. - Make frequent small tasks—JSON formatting, UUID generation—into instantly callable skills. - Accumulate debugging playbooks as skills so anyone gets the same quality of execution. - Capture internal conventions and checklists in SKILL.md to onboard new members naturally. ⚠️ Caveats - Only SKILL.md is required, but if you don't clearly state when to use it, the agent may not fire it appropriately. - Global-scope skills affect every project, so keep narrowly useful ones in the workspace scope. - Skills are designed to save context, so scope each one tightly rather than cramming too much in. #Antigravity# #AIcoding#

Antigravity x Insta360 Global Awards 2026 is now live. 🗓️ Submissions open: Now – October 31 Over 100 chances to win and a total prize pool of US$150,000+. 📲 For social media submissions To qualify, entries must include: Main hashtag: #AntigravityAwards# + #Insta360LeicaAwards# For full details: We look forward to seeing your work. Good luck!

# Antigravity Features and Practical Usage 🚀 One API call to spin up an agent that writes code, runs it, and browses the web. That is what the SDK and Managed Agents make real. 📌 Title and Feature URL Title: SDK / Managed Agents URL: 📝 Overview The Antigravity SDK gives you programmatic access to the same agent harness that powers Google's products. Managed Agents, offered as a capability inside the Gemini API, let you spin up an agent in an isolated Linux environment with a single API call—where it reasons, uses tools, and executes code. 🔧 How It Works The SDK and Managed Agents turn agents into embeddable building blocks: - Isolated execution: each interaction creates a dedicated sandbox where code runs separately and safely. - Stateful, persistent environments: an environment can be resumed in follow-up calls with all files and state intact, so multi-turn work continues without reinitializing. - Custom agent definitions: you describe behavior in markdown files and can start fast from templates in the Google AI Studio Playground. - The underlying model is Gemini 3.5 Flash, optimized for concurrent execution, and integrates with the Interactions API and AI Studio. 🛠 Practical Usage - Call Managed Agents through the Gemini API to launch an agent environment in a single call. - Use the SDK to define custom agent behaviors and host them on the infrastructure of your choice. - Build on the custom agent templates in the AI Studio Playground and extend them for your use case. - Resume a prior environment to keep its files and working state across turns. 🎯 Use Cases - Spin up a sandboxed agent in one call to build a deployment-automation agent. - Embed backend workflows that involve code execution, file operations, and web browsing into internal tools. - Use stateful environments to run multi-turn research or build tasks continuously. - Host SDK-defined agents on your own infrastructure to match operational requirements. ⚠️ Caveats - Managed Agents is a Gemini API surface, so it requires an API key and a billing-enabled project. - Because each interaction creates an environment, design with sandbox lifecycle and cost in mind. - Since agents run code and browse the web, do not skip permission scoping and input validation. #Antigravity# #Agents#

# Antigravity Features and Practical Usage 🚀 No browser, no IDE—just a terminal and an agent ready to build. That is the promise of the Antigravity CLI (agy). 📌 Title and Feature URL Title: Antigravity CLI URL: 📝 Overview The Antigravity CLI brings the same agent harness that powers Antigravity 2.0 to the command line. It lets you spin up new agents instantly without a GUI, and it runs fast over SSH on remote servers or inside containers. You drive it with the agy command. 🔧 How It Works agy shares the IDE's core while adding terminal-first extensibility: - AGENTS.md at the repo root supplies shared project instructions, replacing the older .gemini/ convention with a new project-instructions system. - Agent Skills load from a global scope (~/.gemini/antigravity-cli/skills/) and a per-workspace scope (.agents/skills/). - Hooks intercept agent actions at lifecycle stages (before a tool call, after a model call, at loop stop conditions) to automate checks and formatting. - MCP servers can be configured as local processes or remote hosts to reach external data and tools. - Subagents and Plugins are also first-class extensibility features. 🛠 Practical Usage - Install on macOS/Linux with curl -fsSL | bash, or on Windows via PowerShell. - Run agy and authenticate with a Google account or a GCP project on first use. - Run agy inspect to see exactly what context is loaded: AGENTS.md, available skills, plugins, registered hooks, and connected MCP servers. - Use /goal to run autonomously, /grill-me to have the agent ask clarifying questions first, /schedule for recurring runs, and /browser to opt into browser capabilities. 🎯 Use Cases - Run agents directly from the terminal where you cannot open an IDE, such as SSH or remote sessions. - Debug "why isn't my skill firing?" or "which MCP loaded?" by inspecting the live context with agy inspect. - Launch lightweight agent work inside CI containers. - Standardize per-repo conventions by pairing AGENTS.md with shared team skills. ⚠️ Caveats - The default model is Gemini 3.5 Flash (High); preview rate limits are generous, but production workloads need a billing-enabled GCP project. - When skills or hooks misbehave, check the real loaded state with agy inspect before blaming your config. - Note that global and workspace scopes resolve skills from different directories. #Antigravity# #DevTools#