Felo SuperAgent: Run Continuous AI Employee Workflows with Memory in Google Antigravity

Keep long-running agent tasks alive with streaming answers and LiveDoc-backed memory. Learn how Felo SuperAgent extends Google Antigravity's Agent Manager for ongoing research, support, and operating workflows.

Google Antigravity's Agent Manager is a powerful planner. It breaks complex tasks into steps, coordinates tools, and orchestrates your AI employee through structured workflows. But there's a fundamental limitation baked into every agent session: when the context window resets, everything the agent was working on disappears.

Felo SuperAgent solves that problem. It's the skill that transforms Antigravity from a session-bound task runner into a persistent, continuously operating AI employee — one that remembers what it learned, streams answers as it works, and keeps running long after a single conversation would have ended.

The Problem Antigravity Alone Can't Solve

Antigravity's Agent Manager excels at discrete tasks: "research these competitors," "summarize this document," "generate a report." Each task has a clear beginning and end. The agent plans, executes, and delivers — all within one context window.

But real work doesn't always fit into single sessions. Consider these scenarios:

• Ongoing market monitoring — tracking competitor pricing changes across weeks, not minutes
• Multi-phase research projects — gathering sources over days, synthesizing incrementally
• Continuous support workflows — monitoring tickets, escalating issues, maintaining context across customer interactions
• Long-running investigations — security audits, compliance reviews, or technical deep-dives that span hours or days

In each case, the agent needs something Antigravity doesn't provide out of the box: persistent memory across sessions, streaming real-time responses, and the ability to resume work where it left off.

That's exactly what Felo SuperAgent adds.

What Felo SuperAgent Brings to Antigravity

Felo SuperAgent is a skill folder you drop into .agent/skills/ and commit to Git — the same installation pattern as every Felo skill for Antigravity. No configuration, no per-user setup, no infrastructure to manage. The Agent Manager picks it up automatically via SKILL.md semantic routing.

Once installed, it adds three capabilities that transform how your agents operate.

1. Streaming Answers — Real-Time Output, Not Batch Results

Traditional agent workflows are batch-oriented: the agent plans, executes, and then returns the complete result at the end. For simple tasks, that works fine. For complex, multi-step research or analysis, waiting minutes (or longer) for a single response creates friction.

Felo SuperAgent uses Server-Sent Events (SSE) to stream answers as they're generated. Think of it as the difference between downloading a file and streaming video — you get information as it becomes available, not after everything is finished.

This matters for several reasons:

• Interactivity — You can steer the agent mid-task. See a research direction go off-track? Redirect it immediately instead of waiting for the full output.
• Transparency — Watch the agent's reasoning unfold in real-time. Understand how it arrives at conclusions, not just what it concludes.
• Early value — Get actionable insights before the full analysis is complete. The first findings arrive while deeper research continues in the background.

2. LiveDoc-Backed Memory — Persistent Knowledge That Outlives Sessions

The most transformative feature of Felo SuperAgent is its integration with LiveDoc, Felo's persistent knowledge layer. When an agent runs a SuperAgent workflow, it doesn't just produce an answer — it writes its findings, reasoning, and context into a LiveDoc that persists beyond the session.

Here's what that means in practice:

Without SuperAgent: An agent researches your competitor's pricing, delivers a summary, and then forgets everything. Next week, when you ask the same question, the agent starts from scratch — same search, same analysis, same cost.

With SuperAgent: The agent's research lives in a LiveDoc. When a follow-up task references the same topic, the agent retrieves its previous findings and builds on them. It doesn't re-research what it already knows. It starts from where it left off.

This creates a compound intelligence effect: each agent interaction makes the next one faster, smarter, and more grounded in accumulated context.

3. Continuous Workflow Engine — Agents That Keep Running

SuperAgent workflows aren't limited to a single query-response cycle. They support continuous, long-running operations that keep processing in the background.

Imagine an agent that:

1. Monitors your competitors' pricing pages daily
2. Detects when a competitor changes their pricing structure
3. Streams an alert the moment the change is detected
4. Updates a persistent LiveDoc with the new pricing history
5. Maintains a running competitive analysis that grows richer over time

This isn't a hypothetical. It's exactly the kind of workflow SuperAgent enables — an AI employee that doesn't just respond to requests, but actively operates on an ongoing basis.

How It Works: Architecture Overview

Felo SuperAgent sits between Antigravity's Agent Manager and Felo's LiveDoc infrastructure. Here's the flow:

1. Agent Manager triggers SuperAgent — When a task matches the SuperAgent skill's description (continuous workflow, ongoing research, streaming answers), the Agent Manager loads it automatically via SKILL.md semantic routing.

2. SuperAgent establishes an SSE stream — Instead of returning a single batch response, the agent opens a streaming connection that delivers partial results as they're generated.

3. Findings write to LiveDoc — As the agent discovers information, it writes structured findings to a named LiveDoc. This becomes the persistent memory layer.

4. Subsequent tasks retrieve from LiveDoc — When a new agent task references the same topic, it retrieves prior findings from LiveDoc and continues from where the previous session ended.

The result is a stateful agent workflow running on top of Antigravity's stateless Agent Manager — the best of both worlds.

Real-World Use Cases

Competitive Intelligence Dashboard

Set up a SuperAgent workflow that monitors competitor websites, pricing pages, and announcement blogs. The agent:

• Checks designated URLs on a schedule
• Streams changes as they're detected
• Maintains a LiveDoc with a running timeline of competitive moves
• Generates weekly summary reports from accumulated data

Your team gets a continuously updated competitive intelligence feed, powered by an agent that never forgets.

Customer Support Escalation Engine

Configure a SuperAgent workflow that monitors your support queue, identifies patterns in incoming tickets, and maintains context about recurring issues. The agent:

• Processes incoming tickets in real-time
• Streams priority alerts for urgent issues
• Writes resolution patterns to LiveDoc for future reference
• Builds a knowledge base that improves with every ticket resolved

Research Pipeline for Product Development

Use SuperAgent to run ongoing research on technology trends, user feedback, and market signals. The agent:

• Continuously scans sources relevant to your product area
• Streams emerging trends and insights as they appear
• Maintains a structured research repository in LiveDoc
• Produces periodic synthesis reports that compound over time

Installation

Installing Felo SuperAgent for Google Antigravity follows the same pattern as every Felo skill:

# Clone the Felo skills repository
git clone https://github.com/Felo-Inc/felo-skills.git

# Copy the skill to your Antigravity skills folder
cp -r felo-skills/felo-superagent ~/.gemini/antigravity/skills/

Or for team-wide deployment, put it in your project's .agent/skills/ and commit to Git. Every developer's Antigravity instance picks it up automatically on the next pull.

That's it. No configuration files, no API key setup, no infrastructure. The Agent Manager triggers SuperAgent automatically when a task matches its semantic description.

How SuperAgent Fits Into the Felo Skills Ecosystem

Felo SuperAgent doesn't replace other Antigravity skills — it complements them. Here's how it fits:

• Felo Search provides live data that SuperAgent streams into its workflows. SuperAgent is the workflow engine; Search is the data source.
• Felo LiveDoc is the memory layer that SuperAgent reads from and writes to. They're designed to work together — SuperAgent is the primary consumer of LiveDoc in the Antigravity ecosystem.
• Other skills (Felo Slides, Felo Web Fetch, Felo X Search) become data inputs that SuperAgent can orchestrate into multi-step, long-running workflows.

Think of SuperAgent as the conductor of an Antigravity skills orchestra. It doesn't play every instrument — it coordinates the ones that matter for each task, keeps the performance running, and remembers the music for next time.

Why This Matters for Teams

The shift from session-bound agents to persistent, continuously operating AI employees is significant. Here's what changes:

From reactive to proactive. Instead of waiting for you to ask a question, SuperAgent workflows can actively monitor, detect, and alert — turning your AI from a tool you use into a teammate that operates.

From disposable to compounding. Every agent interaction with SuperAgent writes to persistent memory. Knowledge accumulates. The system gets smarter over time.

From isolated to integrated. SuperAgent coordinates multiple skills into unified workflows. Research, analysis, and output generation happen in one continuous process, not as disconnected tasks.

Getting Started

The fastest way to experience SuperAgent is to install it alongside Felo Search and Felo LiveDoc. Start with a simple continuous research workflow:

1. Install all three skills into .agent/skills/
2. Create a LiveDoc for your research topic
3. Ask your Antigravity agent to monitor and report on that topic
4. Watch as the agent streams findings in real-time and writes them to persistent memory
5. Come back tomorrow and ask for an update — the agent remembers everything

From there, expand into more complex workflows: competitive monitoring, support escalation, research pipelines. Each one builds on the same SuperAgent foundation.

Conclusion

Felo SuperAgent transforms Google Antigravity from a capable task planner into a persistent, continuously operating AI employee platform. With streaming answers, LiveDoc-backed memory, and support for long-running workflows, it closes the gap between what agents can do in a single session and what teams actually need from them.

Install it at felo.ai/skills/antigravity, drop it into .agent/skills/, and watch your Antigravity agents evolve from session-bound task runners into persistent, memory-equipped AI teammates.

For the full Felo Skills ecosystem, visit https://felo.ai.

---

This post is also available in 简体中文, 日本語, 한국어, 繁體中文, हिन्दी, Français, العربية, Русский, اردو, Bahasa Indonesia, Deutsch, Tiếng Việt, Türkçe, Italiano, ไทย, Español, বাংলা and Português.