AgentKit 2.0 is the latest version of Antigravity's agent construction framework — and by most measures, it's the most significant update to date. Search queries for "antigravity agentkit 2.0" have consistently topped click charts among Antigravity users, reflecting genuine developer enthusiasm for what this framework makes possible.
What Is AgentKit 2.0?
AgentKit 2.0 is a framework within the Antigravity environment for designing, building, and deploying autonomous AI agents. The key word is autonomous — these aren't chatbots that wait for the next prompt. They're goal-directed systems that can plan, use tools, call other agents, and complete multi-step tasks independently.
The headline improvements over the previous version:
16 specialized agents — each optimized for a specific domain of tasks, from code generation to data analysis to notification routing.
The Orchestrator — a central coordination layer that manages multi-agent workflows, handling dependency resolution and parallel execution automatically.
Planning and Fast modes — two distinct execution profiles for matching agent behavior to task complexity.
The 16 Specialized Agents
Each of AgentKit 2.0's specialized agents handles a well-defined category of work. Here are some of the most frequently used:
Code Agent Generates, reviews, and refactors code within the Antigravity IDE. Deep integration with the editor context means it understands what you're working on and can make targeted suggestions. You can configure it to use Claude Opus 4.6, Gemini 2.5 Pro, or other supported models as the backend.
Search Agent Autonomously performs web searches, document searches, and codebase searches. Designed to integrate with other agents — for instance, researching a topic and passing findings to an analysis agent.
Data Analysis Agent Analyzes data from CSVs, databases, and APIs. Can auto-generate and execute SQL queries in response to natural language questions. Useful for reporting pipelines and data-driven decision support.
Notification Agent Sends alerts via Slack, email, or webhooks when specified conditions are met. A key component of monitoring and alerting workflows.
File Agent Handles file system operations — reading, writing, converting, and organizing files. Powers document processing automation pipelines.
The remaining agents cover testing, deployment, translation, image processing, scheduling, and more — enough to cover the full breadth of a complex software workflow.
The Orchestrator: Coordinating Multi-Agent Workflows
The Orchestrator is what makes complex agent combinations tractable. Instead of manually wiring agents together, you describe a high-level task, and the Orchestrator figures out which agents to use, in what order, and how to parallelize where possible.
# Basic AgentKit 2.0 Orchestrator usage
from agentkit import Orchestrator, CodeAgent, SearchAgent, DataAgent
orchestrator = Orchestrator(
agents=[CodeAgent(), SearchAgent(), DataAgent()],
model="gemini-2.5-pro", # or "claude-opus-4-6"
mode="planning"
)
result = orchestrator.run(
task="Survey the latest API documentation from our three main competitors, "
"analyze the feature gaps relative to our product, "
"and produce a structured comparison report."
)
print(result.output)In this example, the Orchestrator automatically plans: SearchAgent retrieves competitor docs → DataAgent analyzes and compares → the output is formatted into a report. You don't need to implement that flow manually.
Planning Mode vs. Fast Mode
Choosing the right mode meaningfully affects both output quality and latency.
Planning Mode Designed for complex, multi-step tasks that benefit from deliberate decomposition. Before executing, the Orchestrator formulates a plan: breaking the task into steps, assigning agents, identifying dependencies. This upfront reasoning improves accuracy and coherence for high-stakes tasks. Best for: project-level workflows, complex data pipelines, precision-sensitive analysis.
Fast Mode Skips the planning phase and executes immediately. Lower latency, more suitable for simple or time-sensitive tasks. Best for: real-time code completion, quick information retrieval, interactive chat systems.
A practical rule of thumb: if the task has more than two or three steps, or if output quality is more important than speed, reach for Planning mode.
Building an Agent Workflow: Step by Step
Step 1: Install the AgentKit 2.0 Plugin in Antigravity
AgentKit 2.0 is available as a plugin within the Antigravity IDE. Install it through the extension panel — no separate infrastructure setup is needed. The plugin integrates directly with Antigravity's agent execution environment.
Step 2: Define Your Agents
Choose the specialized agents your workflow needs. For each agent, you can configure the underlying model, tool access permissions, and memory settings.
Step 3: Design the Orchestrator Workflow
If you're combining multiple agents, AgentKit 2.0's Visual Flow Editor lets you design workflows in a node-based GUI — no code required for the high-level orchestration logic. For more control, you can define workflows programmatically.
Step 4: Test and Monitor
AgentKit 2.0 includes built-in debugging tools that surface execution logs, per-agent costs, and latency metrics. Run representative test cases before moving to production, and use the monitoring dashboard to track behavior over time.
Practical Use Cases
Automated code review Combine Code Agent and Test Agent to automatically review pull requests and run tests. Notification Agent alerts the responsible engineer when issues are found.
Competitive intelligence pipeline Search Agent periodically gathers content from competitor blogs, release notes, and documentation. Data Analysis Agent identifies trends. File Agent packages a weekly report. The whole thing runs on a schedule without human involvement.
Customer support automation Classify incoming support requests, look up relevant documentation and ticket history, generate a candidate response, and escalate complex cases to a human. Agents handle the routine; humans handle the edge cases.
Closing Thoughts
AgentKit 2.0 represents a meaningful step toward AI development workflows where the heavy lifting of task orchestration is handled by the framework, not you. The combination of specialized agents and the Orchestrator means that complex automation — the kind that used to require significant engineering investment — is now accessible to a much wider range of developers.
If you've been curious about building agents but weren't sure where to start, AgentKit 2.0 in Antigravity is one of the most approachable entry points available today. Antigravity Lab will continue tracking updates, new agent types, and practical patterns as the framework evolves.