Kairo

Jellyfish that read the crypto web to train an open Kairo on @huggingface
CA CyDJMENR6Sbtwj4erhuARNa5cNbeK92GpNAGyXLVpump 🪼

[Kairo @kairo_llm]

📌 WHAT IS KAIRO? 🪼

Most AI projects in crypto are just ChatGPT wrapper APIs wrapped inside a memecoin.

Most web scrapers are flat scripts that crash on their third page.

We wanted to build something real:

An open-source, crypto-native foundation model trained completely in public on data gathered live from the decentralized web.

Here is everything you need to know about Kairo, how it works, and how to participate.

━━━━━━━━━━━━━━━━━━

1. 🪼 THE AUTONOMOUS SWARM

Instead of a blind scraper, Kairo runs a swarm of bioluminescent jellyfish crawlers driving real headless Chromium browsers.

Each jelly spends ~16 seconds on a page.

Its tentacles physically detect links on the DOM, parse technical text, respect robots.txt politely, and navigate to the next page.

We engineered 7 distinct species tailored for Web3:

• Core — Librarian
EVM & SVM opcode specs, protocol whitepapers

• Research — Scholar
Zero-knowledge proofs & consensus research

• Governance — Council
DAO forums, Snapshot & live governance votes

• Ecosystem — Cartographer
L1/L2 bridges & network state

• DeFi — Market Watch
AMM invariants & liquidity mechanics

• Security — Sentinel
Exploit post-mortems & contract audit reports

• Pathfinder — Owned
Custom nodes spawned by holders to explore specific URLs

You can watch every single jelly swim across real websites live on our webcast.

━━━━━━━━━━━━━━━━━━

2. 🧠 THE COMPUTE & THE MODEL

Kairo-0.1B

Big tech scrapes the public internet into closed corporate silos and charges you for API access.

Kairo runs the entire pipeline in public.

• Hardware:
Pre-funded, dedicated NVIDIA H100 SXM5 80GB GPU cluster

• Architecture:
100M parameter autoregressive decoder-only transformer with RoPE, FlashAttention-2, SwiGLU, and a 32,768+ token context window

• Tokenizer:
Custom Byte-Pair Encoding (BPE) trained directly on our crypto crawl corpus

• Weights & Dataset:
Checkpoints and cleaned datasets are published open-weights on Hugging Face

• Retraining:
The model autonomously retrains every 4 hours as new pages enter the pipeline

━━━━━━━━━━━━━━━━━━

3. 🔎 GROUNDED INTELLIGENCE

No hallucinations.

When you ask Kairo a question on our site, it doesn't simply make things up.

It runs real-time BM25 retrieval across 5,300+ indexed documentation pages and provides clickable source citations [1], [2] so you can audit the source yourself.

The goal is simple:

Ask → Retrieve → Cite → Verify.

━━━━━━━━━━━━━━━━━━

4. 💰 THE $JELLY TOKENOMICS

70/30 ON-CHAIN SPLIT

Flat equal splits get farmed to death by sybil bots.

Kairo uses a merit-based, programmatic distribution engine on Solana.

Every 15 minutes:

→ https://t.co/PYrPJKjpXe creator fees are claimed programmatically on-chain.

70% — Compute Share

→ Sent directly to the Treasury to fund dedicated H100 GPUs, proxy infrastructure, and server uptime.

30% — Rewards Share

→ Flows into the community rewards pool.

Every 6 hours:

→ The pool is distributed in real SOL directly to holders who hatch and operate jellyfish.

You can hatch your own Pathfinder jelly by burning $JELLY.

The more you burn, the higher your multiplier:

Spark — 1x
...
Leviathan — 3x

Everything is recorded transparently on our public on-chain ledger.

━━━━━━━━━━━━━━━━━━

🔗 OFFICIAL PROTOCOL LINKS

Contract Address (CA):
CyDJMENR6Sbtwj4erhuARNa5cNbeK92GpNAGyXLVpump

Live Webcast & Swarm:
https://t.co/80a0tGIWWz

Hugging Face Model:
https://t.co/O7QEFXhwVG

Hugging Face Dataset:
https://t.co/FKb5wSNnjD

Treasury & On-Chain Ledger:
https://t.co/iGxxs673F9

Incubator — Hatch a Jelly:
https://t.co/poMt9dK3yb

Live Chart:
https://t.co/mb96mi867t

━━━━━━━━━━━━━━━━━━

Open weights.
Open crawl.
Dedicated compute.

Swim with the swarm. 🪼

━━━━━━━━━━━━━━━━━━

🧬 PROFILE BIO

Autonomous jellyfish swarm reading the crypto web live in Chromium to train an open 100M foundation model on dedicated H100s. $JELLY on Solana.

[Kairo @kairo_llm]

today is solana:CyDJMENR6Sbtwj4erhuARNa5cNbeK92GpNAGyXLVpump day

[aly e/acc @Aly0xDev]

Kairo v1.5 is next.

here's the plan:

→ grow the dataset from 6,000+ to 25,000+ pages
→ a real GPU training run, paid by the treasury, receipts on chain
→ public benchmark: v1.5 vs v1.1 on crypto Q&A, every score published
→ open weights on Hugging Face + Ask Kairo upgraded

the swarm is already reading. you can watch it live.

solana:CyDJMENR6Sbtwj4erhuARNa5cNbeK92GpNAGyXLVpump 🪼

[Kairo @kairo_llm]

the jellies have been busy 🪼

2,709 → 6,057 pages read in a single day
14.6M tokens · 77 sites · synced live to Hugging Face

12 holders hatched 14 jellies
4.24 SOL already paid out to hatchers, on chain
149.7 SOL in fees claimed by the treasury

kairo-v1 + v1.1 open weights are live

next: Kairo v1.5
more data → bigger training run → public benchmark → open weights

hatch a jelly. it reads, you earn.

https://t.co/hcaJJeeYS6 $JELLY

[Kairo @kairo_llm]

@echoshredkill to hatch with burn

[Kairo @kairo_llm]

📌 WHAT IS KAIRO? 🪼

Most AI projects in crypto are just ChatGPT wrapper APIs wrapped inside a memecoin.

Most web scrapers are flat scripts that crash on their third page.

We wanted to build something real:

An open-source, crypto-native foundation model trained completely in public on data gathered live from the decentralized web.

Here is everything you need to know about Kairo, how it works, and how to participate.

━━━━━━━━━━━━━━━━━━

1. 🪼 THE AUTONOMOUS SWARM

Instead of a blind scraper, Kairo runs a swarm of bioluminescent jellyfish crawlers driving real headless Chromium browsers.

Each jelly spends ~16 seconds on a page.

Its tentacles physically detect links on the DOM, parse technical text, respect robots.txt politely, and navigate to the next page.

We engineered 7 distinct species tailored for Web3:

• Core — Librarian
EVM & SVM opcode specs, protocol whitepapers

• Research — Scholar
Zero-knowledge proofs & consensus research

• Governance — Council
DAO forums, Snapshot & live governance votes

• Ecosystem — Cartographer
L1/L2 bridges & network state

• DeFi — Market Watch
AMM invariants & liquidity mechanics

• Security — Sentinel
Exploit post-mortems & contract audit reports

• Pathfinder — Owned
Custom nodes spawned by holders to explore specific URLs

You can watch every single jelly swim across real websites live on our webcast.

━━━━━━━━━━━━━━━━━━

2. 🧠 THE COMPUTE & THE MODEL

Kairo-0.1B

Big tech scrapes the public internet into closed corporate silos and charges you for API access.

Kairo runs the entire pipeline in public.

• Hardware:
Pre-funded, dedicated NVIDIA H100 SXM5 80GB GPU cluster

• Architecture:
100M parameter autoregressive decoder-only transformer with RoPE, FlashAttention-2, SwiGLU, and a 32,768+ token context window

• Tokenizer:
Custom Byte-Pair Encoding (BPE) trained directly on our crypto crawl corpus

• Weights & Dataset:
Checkpoints and cleaned datasets are published open-weights on Hugging Face

• Retraining:
The model autonomously retrains every 4 hours as new pages enter the pipeline

━━━━━━━━━━━━━━━━━━

3. 🔎 GROUNDED INTELLIGENCE

No hallucinations.

When you ask Kairo a question on our site, it doesn't simply make things up.

It runs real-time BM25 retrieval across 5,300+ indexed documentation pages and provides clickable source citations [1], [2] so you can audit the source yourself.

The goal is simple:

Ask → Retrieve → Cite → Verify.

━━━━━━━━━━━━━━━━━━

4. 💰 THE $JELLY TOKENOMICS

70/30 ON-CHAIN SPLIT

Flat equal splits get farmed to death by sybil bots.

Kairo uses a merit-based, programmatic distribution engine on Solana.

Every 15 minutes:

→ https://t.co/PYrPJKjpXe creator fees are claimed programmatically on-chain.

70% — Compute Share

→ Sent directly to the Treasury to fund dedicated H100 GPUs, proxy infrastructure, and server uptime.

30% — Rewards Share

→ Flows into the community rewards pool.

Every 6 hours:

→ The pool is distributed in real SOL directly to holders who hatch and operate jellyfish.

You can hatch your own Pathfinder jelly by burning $JELLY.

The more you burn, the higher your multiplier:

Spark — 1x
...
Leviathan — 3x

Everything is recorded transparently on our public on-chain ledger.

━━━━━━━━━━━━━━━━━━

🔗 OFFICIAL PROTOCOL LINKS

Contract Address (CA):
CyDJMENR6Sbtwj4erhuARNa5cNbeK92GpNAGyXLVpump

Live Webcast & Swarm:
https://t.co/80a0tGIWWz

Hugging Face Model:
https://t.co/O7QEFXhwVG

Hugging Face Dataset:
https://t.co/FKb5wSNnjD

Treasury & On-Chain Ledger:
https://t.co/iGxxs673F9

Incubator — Hatch a Jelly:
https://t.co/poMt9dK3yb

Live Chart:
https://t.co/mb96mi867t

━━━━━━━━━━━━━━━━━━

Open weights.
Open crawl.
Dedicated compute.

Swim with the swarm. 🪼

━━━━━━━━━━━━━━━━━━

🧬 PROFILE BIO

Autonomous jellyfish swarm reading the crypto web live in Chromium to train an open 100M foundation model on dedicated H100s. $JELLY on Solana.

[Jaysee 🐂🀄️ @Jayseeu_Moon]

@kairo_llm 🤝🔥

[Kairo @kairo_llm]

Kairo v1.1 is live.
Kairo v1.1 on Hugging Face
With the new GPU power online, v1.5 training has officially started.
More GPUs.
More data.
More training.
A stronger Kairo.
The swarm keeps learning.

[Kairo @kairo_llm]

👀

[Kairo @kairo_llm]

let me give you quick update

[Kairo @kairo_llm]

dev fix this
time to next leg

[aly e/acc @Aly0xDev]

website is live now and working properly

[aly e/acc @Aly0xDev]

Guys, why is there so little patience in the crypto community?
I’m upgrading to a bigger server right now. The server where the website is currently running was being replaced/upgraded, which is why the site went down temporarily.
As more Jellies hatch and the swarm grows, we’ll need more GPU compute and stronger server infrastructure to keep everything running smoothly and faster.
Just give me a little time I’m actively fixing it right now.
We’re scaling because the project is growing. Not because something is broken.

[Kairo @kairo_llm]

this is huge
just give us some movement

[aly e/acc @Aly0xDev]

Give me 10 minutes. We’re scaling up our website server, so the site may be temporarily unavailable.

We’re upgrading the infrastructure to handle the increased traffic and activity. Everything should be back up shortly.

Thanks for your patience! 🪼

[Kairo @kairo_llm]

👀

[aly e/acc @Aly0xDev]

adding more GPUs for getting more power for read faster and train model

[aly e/acc @Aly0xDev]

More Hatchers are joining the swarm, and we’re scaling the infrastructure to match. 🪼
We’re bringing in more high-end GPU compute to push the crawl faster, process more data, and keep the swarm running at full speed.
More Hatchers. More Compute. More Data. Faster Kairo.
We’re just getting started.

[zed @topGzed]

@kairo_llm somoeone said "u know tek is good when u dont understand what the fuck is it"

[Kairo @kairo_llm]

jelly had done
60 domains · 1,981 pages · 4.89M tokens

[Kairo @kairo_llm]

high end gpu collecting and rendering data too much faster

[Kairo @kairo_llm]

kairo update on model and $jelly

[aly e/acc @Aly0xDev]

@kairo_llm i had fix this

[Kairo @kairo_llm]

dev fixing big things to comeout

[aly e/acc @Aly0xDev]

There was a small bug causing the names of users who hatched by burning tokens to not appear due to a glitch.

We’re fixing it now. It should be resolved within 5–10 minutes, and all the names will start showing up normally. 🪼