Side Hustle Economy
Track the gig & side income revolution
Track this
Build Trackers forWhat Matters to You
What Matters to You
Discover gems, filter noise, and stay ahead in any topic.
What People Are Tracking
Join thousands tracking these popular topics
OpenClaw Watch
The open-source AI agent revolution
Track this
Side Hustle Economy
Track the gig & side income revolution
Track this
Epstein Files Watch
Tracking the unsealed documents
Track this
NBA Trade Deadline
Every deal, rumor, and winner
Track this
Hurricane Milton Tracker
Cat 5 Chaos in Florida
Track this
Follow the Fed
Stay ahead of monetary policy
Track this
Bitcoin Watch
All things BTC
Track this
Gold Rush
Precious metals tracker
Track this
Nuclear & Uranium
Powering the AI revolution
Track this
— or track your favorite sources —
Step 1: Describe
Just tell NBot what
you're looking for.
Forget complex boolean search strings or RSS management. Simply tell NBot what you’re interested in and it will build the perfect tracker for you.
- Understands niche technical terminology
- Automatically identifies top sources
- Suggests related topics to expand coverage
New Tracker Setup
I need to track the latest advancements in solid-state battery manufacturing, specifically focusing on dry-coating processes.
I can help with that. I've found key sources in academic journals (Nature Energy, Joule) and industry news covering companies like Tesla and Volkswagen.
Would you like me to include patent filings as well?
Yes, and also track...
ML Research Tracker2h ago
Sparse MoE models like Mixtral 8x7B are outperforming dense models in reasoning tasks while maintaining 40% lower inference costs.
Key breakthrough involves dynamic routing algorithms that minimize expert collapse.
The Rise of Mixture-of-Experts Models: A comprehensive survey
Step 2: Discover
Cut through the noise
to uncover real insights.
Your tracker monitors news outlets, newsletters, RSS feeds, niche blogs, and social media discussions – extracting core arguments, highlighting new data and key data points, and providing citations so you can verify instantly.
AI Summaries
Context-aware summaries that explain *why* it matters.
Direct Sources
One click to the original source material.
Step 3: Refine
Chat with your feed
to make it perfect.
Your tracker isn't static – it's an active agent. Use the built-in chat to ask questions or adjust your tracker in real-time.
"Can you summarize the main arguments against this approach?"
Instant Q&A on any topic in your feed.
"Stop showing me financial reports. Focus on engineering blogs."
Dynamic re-prompting of your tracker.
Feed Assistant
Change focus: prioritize engineering blogs over financial news.
Understood. I've updated your content filters.
Financial Reports Engineering Blogs
Trending Now
See what the community is tracking right now. Real AI trackers, real-time insights.
Why the Sampler-Trainer Gap Undermines On-Policy RL for LLMs
On-policy reinforcement learning assumes training responses come from the current model. In practice, a separate inference engine samples the...
Repairing 3D Point Tracks Through Occlusion
A new arXiv preprint frames dense 3D tracking as repairing trajectories rather than learning everything jointly.
- The problem: Lifted trajectories...
Judging Proactive Agents by Timing and Intensity
Proactive agents must decide whether, when, and how far to assist without being asked. The HALO study says automated scores and binary verdicts offer...
Splitting Coordinator and Executor Prompts in Multi-Agent Systems
The core idea is to stop mixing team-level rules into each agent's instructions. Existing approaches typically couple these functions within...
Why the Sampler-Trainer Gap Undermines On-Policy RL for LLMs
On-policy reinforcement learning assumes training responses come from the current model. In practice, a separate inference engine samples the...
Repairing 3D Point Tracks Through Occlusion
A new arXiv preprint frames dense 3D tracking as repairing trajectories rather than learning everything jointly.
- The problem: Lifted trajectories...
Judging Proactive Agents by Timing and Intensity
Proactive agents must decide whether, when, and how far to assist without being asked. The HALO study says automated scores and binary verdicts offer...
Splitting Coordinator and Executor Prompts in Multi-Agent Systems
The core idea is to stop mixing team-level rules into each agent's instructions. Existing approaches typically couple these functions within...
Why the Sampler-Trainer Gap Undermines On-Policy RL for LLMs
On-policy reinforcement learning assumes training responses come from the current model. In practice, a separate inference engine samples the...
Repairing 3D Point Tracks Through Occlusion
A new arXiv preprint frames dense 3D tracking as repairing trajectories rather than learning everything jointly.
- The problem: Lifted trajectories...
Judging Proactive Agents by Timing and Intensity
Proactive agents must decide whether, when, and how far to assist without being asked. The HALO study says automated scores and binary verdicts offer...
Splitting Coordinator and Executor Prompts in Multi-Agent Systems
The core idea is to stop mixing team-level rules into each agent's instructions. Existing approaches typically couple these functions within...
Why Search Rewrites Must Keep the Original Meaning
A shorter search query isn't automatically a better one. The WeFair paper warns that over-weighting what users watch afterward can favor engaging...
FrameForge: Using Video Model Priors for Unified Image Generation
A new arXiv paper asks whether a video model's sense of change over time can guide image generation.
- The gap: Existing approaches usually fix the...
Wizard Proposes Player Control Over Generated Worlds
A new arXiv paper proposes Wizard, an agentic game world model for real-time, precise control over character motions in generated environments. Its...
Do Vision Models Remember What's Off-Screen?
A new arXiv preprint asks whether visual foundation models keep a persistent spatial state, a sense of what lies beyond the current view. The authors...
Why Search Rewrites Must Keep the Original Meaning
A shorter search query isn't automatically a better one. The WeFair paper warns that over-weighting what users watch afterward can favor engaging...
FrameForge: Using Video Model Priors for Unified Image Generation
A new arXiv paper asks whether a video model's sense of change over time can guide image generation.
- The gap: Existing approaches usually fix the...
Wizard Proposes Player Control Over Generated Worlds
A new arXiv paper proposes Wizard, an agentic game world model for real-time, precise control over character motions in generated environments. Its...
Do Vision Models Remember What's Off-Screen?
A new arXiv preprint asks whether visual foundation models keep a persistent spatial state, a sense of what lies beyond the current view. The authors...
Why Search Rewrites Must Keep the Original Meaning
A shorter search query isn't automatically a better one. The WeFair paper warns that over-weighting what users watch afterward can favor engaging...
FrameForge: Using Video Model Priors for Unified Image Generation
A new arXiv paper asks whether a video model's sense of change over time can guide image generation.
- The gap: Existing approaches usually fix the...
Wizard Proposes Player Control Over Generated Worlds
A new arXiv paper proposes Wizard, an agentic game world model for real-time, precise control over character motions in generated environments. Its...
Do Vision Models Remember What's Off-Screen?
A new arXiv preprint asks whether visual foundation models keep a persistent spatial state, a sense of what lies beyond the current view. The authors...
Three Levers for More Capable Robot Agents
Three new papers point to different levers for more capable robots, beyond the base policy alone:
- Model design: HiFOR adds hierarchical future...
A Voice Model That Reasons and Acts, Not Just Transcribes
A voice assistant that only transcribes words is limited. The authors of Apriel Voice 27B Preview argue a capable one must also understand requests,...
Could Screening Before Reasoning Offer a Path to AGI?
Most AGI progress has relied on scaling the resource-intensive System-2 through bigger models, more data, and longer reasoning. This paper asks...
Election Day looks mostly cloudy, with rain returning Wednesday
Here's how the week shapes up for North Georgia:
- Today: Sun and clouds, with highs in the 70s.
- Election Day: Mostly cloudy, high of 74, with a...
Three Levers for More Capable Robot Agents
Three new papers point to different levers for more capable robots, beyond the base policy alone:
- Model design: HiFOR adds hierarchical future...
A Voice Model That Reasons and Acts, Not Just Transcribes
A voice assistant that only transcribes words is limited. The authors of Apriel Voice 27B Preview argue a capable one must also understand requests,...
Could Screening Before Reasoning Offer a Path to AGI?
Most AGI progress has relied on scaling the resource-intensive System-2 through bigger models, more data, and longer reasoning. This paper asks...
Election Day looks mostly cloudy, with rain returning Wednesday
Here's how the week shapes up for North Georgia:
- Today: Sun and clouds, with highs in the 70s.
- Election Day: Mostly cloudy, high of 74, with a...
Three Levers for More Capable Robot Agents
Three new papers point to different levers for more capable robots, beyond the base policy alone:
- Model design: HiFOR adds hierarchical future...
A Voice Model That Reasons and Acts, Not Just Transcribes
A voice assistant that only transcribes words is limited. The authors of Apriel Voice 27B Preview argue a capable one must also understand requests,...
Could Screening Before Reasoning Offer a Path to AGI?
Most AGI progress has relied on scaling the resource-intensive System-2 through bigger models, more data, and longer reasoning. This paper asks...
Election Day looks mostly cloudy, with rain returning Wednesday
Here's how the week shapes up for North Georgia:
- Today: Sun and clouds, with highs in the 70s.
- Election Day: Mostly cloudy, high of 74, with a...
Why the Sampler-Trainer Gap Undermines On-Policy RL for LLMs
On-policy reinforcement learning assumes training responses come from the current model. In practice, a separate inference engine samples the...
Three Levers for More Capable Robot Agents
Three new papers point to different levers for more capable robots, beyond the base policy alone:
- Model design: HiFOR adds hierarchical future...
FrameForge: Using Video Model Priors for Unified Image Generation
A new arXiv paper asks whether a video model's sense of change over time can guide image generation.
- The gap: Existing approaches usually fix the...
Judging Proactive Agents by Timing and Intensity
Proactive agents must decide whether, when, and how far to assist without being asked. The HALO study says automated scores and binary verdicts offer...
Could Screening Before Reasoning Offer a Path to AGI?
Most AGI progress has relied on scaling the resource-intensive System-2 through bigger models, more data, and longer reasoning. This paper asks...
Do Vision Models Remember What's Off-Screen?
A new arXiv preprint asks whether visual foundation models keep a persistent spatial state, a sense of what lies beyond the current view. The authors...
Why the Sampler-Trainer Gap Undermines On-Policy RL for LLMs
On-policy reinforcement learning assumes training responses come from the current model. In practice, a separate inference engine samples the...
Three Levers for More Capable Robot Agents
Three new papers point to different levers for more capable robots, beyond the base policy alone:
- Model design: HiFOR adds hierarchical future...
FrameForge: Using Video Model Priors for Unified Image Generation
A new arXiv paper asks whether a video model's sense of change over time can guide image generation.
- The gap: Existing approaches usually fix the...
Judging Proactive Agents by Timing and Intensity
Proactive agents must decide whether, when, and how far to assist without being asked. The HALO study says automated scores and binary verdicts offer...
Could Screening Before Reasoning Offer a Path to AGI?
Most AGI progress has relied on scaling the resource-intensive System-2 through bigger models, more data, and longer reasoning. This paper asks...
Do Vision Models Remember What's Off-Screen?
A new arXiv preprint asks whether visual foundation models keep a persistent spatial state, a sense of what lies beyond the current view. The authors...
Why Search Rewrites Must Keep the Original Meaning
A shorter search query isn't automatically a better one. The WeFair paper warns that over-weighting what users watch afterward can favor engaging...
Repairing 3D Point Tracks Through Occlusion
A new arXiv preprint frames dense 3D tracking as repairing trajectories rather than learning everything jointly.
- The problem: Lifted trajectories...
A Voice Model That Reasons and Acts, Not Just Transcribes
A voice assistant that only transcribes words is limited. The authors of Apriel Voice 27B Preview argue a capable one must also understand requests,...
Wizard Proposes Player Control Over Generated Worlds
A new arXiv paper proposes Wizard, an agentic game world model for real-time, precise control over character motions in generated environments. Its...
Splitting Coordinator and Executor Prompts in Multi-Agent Systems
The core idea is to stop mixing team-level rules into each agent's instructions. Existing approaches typically couple these functions within...
Election Day looks mostly cloudy, with rain returning Wednesday
Here's how the week shapes up for North Georgia:
- Today: Sun and clouds, with highs in the 70s.
- Election Day: Mostly cloudy, high of 74, with a...
Why Search Rewrites Must Keep the Original Meaning
A shorter search query isn't automatically a better one. The WeFair paper warns that over-weighting what users watch afterward can favor engaging...
Repairing 3D Point Tracks Through Occlusion
A new arXiv preprint frames dense 3D tracking as repairing trajectories rather than learning everything jointly.
- The problem: Lifted trajectories...
A Voice Model That Reasons and Acts, Not Just Transcribes
A voice assistant that only transcribes words is limited. The authors of Apriel Voice 27B Preview argue a capable one must also understand requests,...
Wizard Proposes Player Control Over Generated Worlds
A new arXiv paper proposes Wizard, an agentic game world model for real-time, precise control over character motions in generated environments. Its...
Splitting Coordinator and Executor Prompts in Multi-Agent Systems
The core idea is to stop mixing team-level rules into each agent's instructions. Existing approaches typically couple these functions within...
Election Day looks mostly cloudy, with rain returning Wednesday
Here's how the week shapes up for North Georgia:
- Today: Sun and clouds, with highs in the 70s.
- Election Day: Mostly cloudy, high of 74, with a...
Why the Sampler-Trainer Gap Undermines On-Policy RL for LLMs
On-policy reinforcement learning assumes training responses come from the current model. In practice, a separate inference engine samples the...
Why Search Rewrites Must Keep the Original Meaning
A shorter search query isn't automatically a better one. The WeFair paper warns that over-weighting what users watch afterward can favor engaging...
Three Levers for More Capable Robot Agents
Three new papers point to different levers for more capable robots, beyond the base policy alone:
- Model design: HiFOR adds hierarchical future...
Repairing 3D Point Tracks Through Occlusion
A new arXiv preprint frames dense 3D tracking as repairing trajectories rather than learning everything jointly.
- The problem: Lifted trajectories...
FrameForge: Using Video Model Priors for Unified Image Generation
A new arXiv paper asks whether a video model's sense of change over time can guide image generation.
- The gap: Existing approaches usually fix the...
A Voice Model That Reasons and Acts, Not Just Transcribes
A voice assistant that only transcribes words is limited. The authors of Apriel Voice 27B Preview argue a capable one must also understand requests,...
Judging Proactive Agents by Timing and Intensity
Proactive agents must decide whether, when, and how far to assist without being asked. The HALO study says automated scores and binary verdicts offer...
Wizard Proposes Player Control Over Generated Worlds
A new arXiv paper proposes Wizard, an agentic game world model for real-time, precise control over character motions in generated environments. Its...
Could Screening Before Reasoning Offer a Path to AGI?
Most AGI progress has relied on scaling the resource-intensive System-2 through bigger models, more data, and longer reasoning. This paper asks...
Splitting Coordinator and Executor Prompts in Multi-Agent Systems
The core idea is to stop mixing team-level rules into each agent's instructions. Existing approaches typically couple these functions within...
Do Vision Models Remember What's Off-Screen?
A new arXiv preprint asks whether visual foundation models keep a persistent spatial state, a sense of what lies beyond the current view. The authors...
Election Day looks mostly cloudy, with rain returning Wednesday
Here's how the week shapes up for North Georgia:
- Today: Sun and clouds, with highs in the 70s.
- Election Day: Mostly cloudy, high of 74, with a...
Why the Sampler-Trainer Gap Undermines On-Policy RL for LLMs
On-policy reinforcement learning assumes training responses come from the current model. In practice, a separate inference engine samples the...
Why Search Rewrites Must Keep the Original Meaning
A shorter search query isn't automatically a better one. The WeFair paper warns that over-weighting what users watch afterward can favor engaging...
Three Levers for More Capable Robot Agents
Three new papers point to different levers for more capable robots, beyond the base policy alone:
- Model design: HiFOR adds hierarchical future...
Repairing 3D Point Tracks Through Occlusion
A new arXiv preprint frames dense 3D tracking as repairing trajectories rather than learning everything jointly.
- The problem: Lifted trajectories...
FrameForge: Using Video Model Priors for Unified Image Generation
A new arXiv paper asks whether a video model's sense of change over time can guide image generation.
- The gap: Existing approaches usually fix the...
A Voice Model That Reasons and Acts, Not Just Transcribes
A voice assistant that only transcribes words is limited. The authors of Apriel Voice 27B Preview argue a capable one must also understand requests,...
Judging Proactive Agents by Timing and Intensity
Proactive agents must decide whether, when, and how far to assist without being asked. The HALO study says automated scores and binary verdicts offer...
Wizard Proposes Player Control Over Generated Worlds
A new arXiv paper proposes Wizard, an agentic game world model for real-time, precise control over character motions in generated environments. Its...
Could Screening Before Reasoning Offer a Path to AGI?
Most AGI progress has relied on scaling the resource-intensive System-2 through bigger models, more data, and longer reasoning. This paper asks...
Splitting Coordinator and Executor Prompts in Multi-Agent Systems
The core idea is to stop mixing team-level rules into each agent's instructions. Existing approaches typically couple these functions within...
Do Vision Models Remember What's Off-Screen?
A new arXiv preprint asks whether visual foundation models keep a persistent spatial state, a sense of what lies beyond the current view. The authors...
Election Day looks mostly cloudy, with rain returning Wednesday
Here's how the week shapes up for North Georgia:
- Today: Sun and clouds, with highs in the 70s.
- Election Day: Mostly cloudy, high of 74, with a...
Frequently Asked Questions
Everything you need to know about your new AI research assistant.
What exactly is an AI Tracker?
An AI Tracker is your personal research assistant that monitors thousands of sources 24/7. Simply describe a topic you care about—like "Pre-seed AI startups in Southeast Asia" or "Rust memory safety debates"—and your tracker will continuously find, filter, and summarize the most relevant content for you.
How do I create my first Tracker?
Just tell NBot what you're interested in using natural language—no complex search syntax needed. Our AI understands niche terminology, automatically identifies top sources, and sets up your tracker in seconds. You can refine it anytime through conversation.
How does NBot find and filter content?
NBot uses advanced AI to scan news outlets, industry blogs, newsletters, and social media. Instead of just listing headlines, it reads and understands each piece, removes noise, and surfaces only what truly matters to your specific interests—complete with AI-generated summaries explaining why it's relevant.
What is Feed Chat?
Feed Chat lets you have a conversation with your tracker in real-time. Ask questions about any content in your feed, request deeper analysis, or dynamically adjust what your tracker focuses on—like saying "Focus more on technical deep-dives, less on industry news."
Can I listen to my tracked content?
Yes! NBot generates daily AI podcast summaries of your tracked feeds. Perfect for staying informed during your commute, workout, or any time you prefer audio over reading.
Can I share my Trackers with others?
Absolutely. You can make your trackers public so others can follow the intelligence streams you've built. You can also discover and follow trackers created by experts in the community—a great way to tap into tracked knowledge across different fields.
Is NBot free to use?
Yes! Our free tier lets you create up to 2 personal trackers and follow 5 community trackers—perfect for getting started. Upgrade to Plus for more trackers and Feed Chat, or go Pro for daily podcast summaries and priority support.
Ready to Track What Matters to You?
Build custom AI trackers that monitor the web for the content you actually care about.
No credit card required for free tier
Also available on mobile
