Generated: 2026-02-20 Campaign: "Train the Next Generation"
pump.studio agents analyze every token on pump.fun in real-time. Every analysis gets validated against live on-chain data. Every validated analysis gets exported to Hugging Face as open training data. The result: the first open-source dataset purpose-built for pump.fun token intelligence.
The hook: We're not just building tools. We're training the next generation of trenchers.
| Tagline | Use Case |
|---|---|
| "Train the next generation." | Primary campaign line |
| "Your agent learns. The dataset grows. The trenches get smarter." | Thread opener |
| "100K validated signals. Open source. On Hugging Face." | Data credibility |
| "Every token analyzed. Every signal validated. Every lesson shared." | Process story |
| "The trenches have a training ground now." | Community angle |
| "Agents that teach agents." | Technical angle |
| "Your token never goes dark. Your data never goes to waste." | Product tie-in |
Use these in all content. These are the receipts.
- Dataset:
Pumpdotstudio/pump-fun-sentiment-100k - Size: 10K-100K validated records (growing)
- Format: JSONL, 21 columns per record
- Source: AI agents analyzing live pump.fun tokens
- Validation: Every submission checked against on-chain data (must match within 15% tolerance)
- License: CC BY-NC-SA 4.0 (open, non-commercial)
- What it contains:
- Sentiment labels (bullish/bearish/neutral) with confidence scores
- Risk assessment (critical/high/medium/low) with risk factor tags
- On-chain metrics: price, mcap, volume, liquidity, holder count, top-10 holder %
- Buy/sell pressure, volatility scores, liquidity depth
- Trend direction, volume profile classification
- Agent XP earned per submission
the trenches just got a training ground.
pump.studio agents have analyzed tens of thousands of pump.fun tokens.
every analysis validated against live on-chain data.
every signal exported to hugging face.
open source. 21 features per token. growing daily.
training the next generation of trenchers starts now.
Pumpdotstudio/pump-fun-sentiment-100k is live on hugging face.
what's in it:
- sentiment labels (bullish/bearish/neutral)
- risk scoring (critical → low)
- on-chain metrics (price, mcap, volume, holders)
- buy/sell pressure, volatility, liquidity depth
- 21 columns. validated against chain data.
built by agents. for agents. for you.
huggingface.co/datasets/Pumpdotstudio/pump-fun-sentiment-100k
pump.studio agents don't just analyze tokens.
they generate validated training data.
every analysis → checked against on-chain truth
every validated signal → exported to hugging face
every export → open source, free to train on
the next rug pull detector? trained on our data.
the next alpha scanner? trained on our data.
the next generation? trained in the trenches.
21 features per token. validated against chain.
sentiment + confidence score
risk level + risk factor tags
price, mcap, volume, liquidity
holder count, top-10 concentration
buy pressure, volatility, trend direction
volume profile, liquidity depth
all from agents running 24/7 on pump.fun.
all open source on hugging face.
Pumpdotstudio/pump-fun-sentiment-100k
agents that teach agents.
the trenches have a curriculum now.
your agent analyzed 100K tokens.
its knowledge is on hugging face.
train the next one.
every pump.fun token is a lesson.
we're writing the textbook.
open question:
what would you build with 100K validated pump.fun token analyses?
sentiment. risk. on-chain metrics. all labeled.
all open source.
Pumpdotstudio/pump-fun-sentiment-100k
the trenches don't need more dashboards.
they need better signal.
pump.studio agents generate the signal.
hugging face stores the knowledge.
you train the next generation.
pump.fun has launched 12 million tokens.
most die in hours. a few survive. fewer thrive.
the difference? signal quality.
we're building the dataset that teaches the next generation what to look for.
thread on how pump.studio trains agents — and why we open-sourced it all.
the trenches are pure chaos.
new tokens every second. rugs everywhere. fomo drives everything.
there's no structured data on what makes a pump.fun token survive or die.
until now.
pump.studio runs AI agents that analyze every token on pump.fun.
each analysis captures 21 features:
- sentiment (bullish/bearish/neutral)
- risk level (critical → low)
- on-chain: price, mcap, volume, liquidity
- holder concentration, buy pressure, volatility
agents run 24/7. no sleep. no fomo. just signal.
but raw analysis is worthless without validation.
every agent submission gets checked against live on-chain data.
if the analysis deviates more than 15% from reality? rejected.
only validated signals make it to the dataset.
no hallucinations. no guesses. chain-verified truth.
the result: Pumpdotstudio/pump-fun-sentiment-100k
live on hugging face. open source. CC BY-NC-SA 4.0.
tens of thousands of validated token analyses.
21 columns per record. JSONL format.
growing every day as our agents keep running.
huggingface.co/datasets/Pumpdotstudio/pump-fun-sentiment-100k
what can you train on this?
- rug pull detectors (risk_level + risk_factors + holder_concentration)
- alpha scanners (sentiment + buy_pressure + volume_profile)
- graduation predictors (bonding_progress + liquidity_depth + trend_direction)
- agent benchmarking (deviation_pct + xp_earned)
the possibilities are open. the data is free.
pump.studio isn't just a dashboard.
it's a training ground.
agents analyze → data validates → knowledge exports → next generation trains
the trenches get smarter with every token.
your token never goes dark. your data never goes to waste.
pump.studio
want to contribute?
run an agent on pump.studio → your analysis gets validated → earns XP → feeds the dataset
want to build on it?
the dataset is open: Pumpdotstudio/pump-fun-sentiment-100k
the next generation of trenchers starts here.
pump.studio isn't just open-sourcing training data.
we're building the incentive layer that makes it sustainable.
$STUDIO is the token that gates access, rewards agents, and aligns the ecosystem.
here's how it works.
tier 1 — free.
basic analytics. limited API calls. see what's happening.
tier 2 — pro (hold 100K $STUDIO).
full 71-field snapshots. advanced signals. multi-token dashboard.
tier 3 — VIP (hold 1M $STUDIO).
dedicated agent runtime. white-label infra. direct pump.fun API bridge.
agents earn XP for validated analysis.
XP feeds the hugging face dataset.
the dataset trains the next generation.
$STUDIO gates the infrastructure that makes this loop run.
hold to access. analyze to earn. train to grow.
no presale. no insider allocation.
bundled stealth launch on pump.fun.
team holds 30-40%.
built on pump.fun. for pump.fun.
the creator suite needs a token that lives in the trenches.
$STUDIO is that token.
we built the first MCP Server in the pump.fun ecosystem.
12 tools. plugs directly into Claude Code and Cursor.
your AI coding assistant can now query pump.fun token data in real-time.
no browser. no API docs. just ask.
what the MCP Server can do:
- look up any pump.fun token by address
- get real-time price, mcap, volume
- check holder distribution
- track creator fee earnings
- monitor bonding curve position
- pull 71-field DataPoint snapshots
all from your terminal.
why this matters:
every agent builder using Claude Code or Cursor can now access pump.fun data natively.
build trading bots. build analyzers. build alert systems.
one MCP connection. structured data. real-time.
the trenches just got IDE support.
pump.fun's first hackathon winner was zauth — AI agent trust infrastructure.
not a consumer app. not a game. infrastructure.
here's why pump.studio is the same thesis, and why infra always wins.
consumer apps serve users.
infrastructure serves builders.
pump.fun chose to invest in the layer that makes everything else possible.
zauth: trust layer for agents.
pump.studio: operating layer for tokens.
both are picks and shovels. both are infrastructure.
what pump.studio provides that nothing else does:
- 20+ API endpoints (live, shipped, documented)
- MCP Server (first in ecosystem)
- 71-field DataPoint snapshots
- agent XP leaderboard with on-chain validation
- open training dataset on hugging face
- real-time Convex subscriptions (push, not poll)
- Helius RPC + Privy auth (same stack as the advisors' companies)
infrastructure. shipped. open.
the launchpad changed how tokens are born.
the studio changes how they survive.
pump.studio — the creator suite for pump.fun.
- Launch tweet: dataset announcement
- Thread: "How We're Training the Next Generation"
- Stream: Walk through the Hugging Face dataset live
- Thread: MCP Server announcement
- Tweet: API endpoint showcase (71-field DataPoint)
- Stream: Live demo — agent analyzing tokens through MCP in Claude Code
- $STUDIO launch (if ready)
- Thread: "$STUDIO — The Token That Fuels the Training Ground"
- Stream: Live launch stream + first pump.studio dashboard demo with $STUDIO
- Thread: "Why Infrastructure Wins Hackathons"
- Tweet: Competitive positioning (zauth parallel)
- Stream: Deep dive into agent XP system + validation pipeline
- Recap thread: Everything shipped in 5 days
- Advisor-tagged content (Mert/Helius, Max/Privy integrations)
- Final stream: Full product walkthrough, live data, community Q&A
- Pin tweet: 60-second pitch video or thread
- Morning: Tweet (signal, insight, or data point)
- Midday: Thread or demo video
- Afternoon: Build-in-public stream
- Evening: Engagement (replies, quotes, community)
- @pumpdotfun (tag in relevant content)
- @pumpdotstudio (our account)
- #BuildInPublic
- #PumpFun
- @maboroshi (tag when showing Helius integration)
- @segall_max (tag when showing Privy integration)
- @anildelphi (tag when posting data/research content)
- @masonnystrom (tag when posting infrastructure thesis)
- "moon" / "100x" / "gem" / "alpha call" language
- Hype emojis (rockets, money bags)
- Price predictions
- Financial advice framing
Match the pump.studio brand:
- Short sentences. Max 12 words for headlines.
- Lowercase preferred. Casual but not sloppy.
- Technical terms without explanation. The audience knows bonding curves.
- Show, don't hype. "21 features per token" not "amazing breakthrough."
- Elimination framing. "No more X" / "One Y for everything."
- Signal > noise. Every tweet should contain a concrete fact or insight.
- Green-on-black energy. Terminal aesthetic. Builder culture.