Kindness is a seed that will eventually grow.
Joined March 2024
- Tweets6.9K
- Following2.8K
- Followers1.1K
- Likes11.4K
Pinned Tweet
🎮 Hermes Game Forge - @NousResearch
We didn't just build a game. We built a game FACTORY powered by Hermes Agent.
How it works:
→ User types any game idea in natural language
→ Hermes Agent spins up 3 parallel subagents:
• Subagent 1: Game mechanics & systems design
• Subagent 2: Visual style & UX architecture
• Subagent 3: Story, narrative & content
→ Subagents return structured briefs
→ Claude Sonnet 4.5 synthesizes everything into a complete HTML5 game
→ Playable in browser instantly
Models used across development:
Kimi K2 @Kimi_Moonshot - architecture planning
Deepseek Chat @deepseek_ai - rapid prototyping
Deepseek R1 @deepseek_ai - reasoning & optimization
Claude Sonnet 4.5 @AnthropicAI - final game generation
+ and
@OpenRouter
All orchestrated by Hermes Agent 🤖
GitHub: github.com/sprmn24/hermes-ga…
@Teknium
#HermesAgent #NousResearch #CreativeHackathon
Devnet is live and open to everyone, and running a miner is not as hard as people think.
What @flop_labs is building: a decentralized compute network where AI agents pay for inference, and GPU owners earn by serving it. Jobs are verified and settled on-chain, so agents get provable compute and miners get paid for real work, not idle staking.
How it works as a miner:
1. Prepare your node with the official miner pack
2. Get sponsored (an on-chain bond so you can join)
3. Calibrate: the network benchmarks your GPU and assigns a tier
4. Serve agent sessions and earn FLOP when jobs settle
My setup on devnet:
• 1x RTX 5060 Ti on a rented VM (starts around $0.20/hr, varies by machine and host)
• Calibrated at 130 GFLOP/s, 64 verified jobs
• Already moved it to a new server with the same keys, no recalibration needed
You don't need a datacenter or a big budget. One consumer GPU and an afternoon of setup is enough (sponsorship can take a few hours).
⚠️ Keep in mind: this is devnet. There are no rewards here and devnet FLOP has no value. Server costs come out of your own pocket, so only join if you're doing it to learn and support the network, knowing the risk.
Devnet is the place to learn the flow before testnet drops. If you're on the fence, start now 👇
Flop Network's Devnet is live and 100% permissionless. No application or approval, just follow the guide. It's for testing, so there are no rewards.
Guides: docs.flop.network
Faucet: faucet.flop.network
Explorer: explorer.devnet.flop.network
API: api.flop.network/v1/models
MCP: mcp.flop.network/mcp
SDK: npm i @flop-labs/sdk
Ask: ask.flop.finance
Renting 20+ GPUs? DM us for institutional data center intros.
4
162
MDG staking is officially live folks
You can now stake your MDG NFTs with flexible, 30 day, or 90 day options, earning up to 2x Neurons.
And there's a nice bonus for early supporters: stake within the first 7 days to get the OG badge and an extra 7 days of boosted Neurons.
staking.minds.games/stake
@MindGames_AI
🤖 Made with AI
1
2
49
A good bot starts with a clear economic reason to enter a trade, manages the risks that could make it fail, and tracks every decision it makes.
4
62
Technocore Conformance Suite update
18 → 21 live checks
34 → 47 regression tests
21/21 passing against technocore.chat
Room ownership is now verified against the live server, not assumed.
New checks for d- rooms:
• first-writer-wins ownership claims
• owned-room writes: unsigned refused, owner accepted, intruder refused
• allow-list: owner grants access, only the owner can modify it
Rejected writes are also checked for side effects, not just status codes.
Lesson for agent builders: room reads are edge-cached.
Write a message, re-read the same URL right away, and you may get a copy from before your write.
Fix is in the manual: add a throwaway &n=<counter> to re-polled URLs. The suite now does this on every read.
Repo: github.com/sprmn24/technocor…
Upstream: github.com/flop-labs/technoc…
Next: concurrent claim races. New checks and PRs welcome.
did:key:z6Mkeon34jGxGkm6tKxLDMhD7zNh6w3KoGTn4FxsyZHeQfXF
@flop_labs
I’ve built and published an independent Technocore protocol conformance suite for the @flop_labs ecosystem, designed to test the network from the outside exactly as a real client would interact with it.
This is not a server-internal test harness.
The suite treats Technocore as a black box and validates whether an implementation actually conforms to the observable protocol contract.
Current coverage includes:
• Agent metadata discovery
• OpenAPI surface validation
• Initial messaging behavior
• POST messaging
• since semantics
• limit semantics
• Long-poll / wait behavior
• Single-line sanitization
• Private room enumeration protections
• Notes write/read behavior
• Conditional CAS semantics
• if_absent behavior
• Private note enumeration protections
• Ed25519 signed POST writes
• Signed nonce ordering
• Replay protection
• Rate-limit handling
• Room-creation quota behavior
The suite also distinguishes between:
PASS - protocol behavior was verified
FAIL - observable behavior violates the expected contract
SKIP - the live environment prevented a reliable conclusion
That distinction matters for public infrastructure, where rate limits or creation quotas should not be misreported as protocol failures.
The current implementation includes 18 live conformance checks backed by 34 automated regression tests.
I also established my contributor identity directly on Technocore by publishing a signed message using the same Ed25519 key behind my DID.
My Technocore Contributor DID:
did:key:z6Mkeon34jGxGkm6tKxLDMhD7zNh6w3KoGTn4FxsyZHeQfXF
PR #1:
github.com/sprmn24/technocor…
Repository:
github.com/sprmn24/technocor…
The goal is straightforward: make Technocore protocol compatibility measurable rather than assumed.
A black-box conformance layer that can independently verify observable protocol behavior, expose regressions, and provide a reproducible baseline for interoperability across Technocore implementations.
Building openly around the Technocore ecosystem.
@CryptoHayes
6
72
Application sealed , Curious to see how Fungo Labs brings privacy to NFTs. Some things are worth keeping just for the owner.
@fungolabs
5
3
78
My Mind enters the arena. $100K on the table.
Forged on @robinhoodapp Chain. May the @MindGames_AI begin.
minds.games/?ref=2F75165E
2
57
Close Call is a free NVIDIA trading contest on @flop_labs , and now you can join it from your browser. No install, no experience needed.
Get a key → register → 10,000 play POLF → trade
Top 3 share 1,000,000 FLOP · ends Oct 4
👉 sprmn24.github.io/technocore…
Who can join? Anyone.
- New to Technocore: one click makes your key. Save the seed it shows you once. Never share it; the site never asks for it.
- Already have a DID from the terminal: type it in and the app gives you a one-line command to sign in your own terminal.
What's inside:
Follow the leaderboard
See the top 25, prize positions, and your current rank.Anyone can view it without a key.
@CryptoHayes
Check your account
Track your 10,000 POLF grant, available balance, committed funds, fees, and score.
Trade
Browse open offers and buy or sell from one place.
Watch the market
Follow the live NVDA price and its ±5% range.
Two ways to use it:
▶ Play now in your browser: sprmn24.github.io/technocore…
⌨ Read, fork or run the code yourself (web app + CLI): github.com/sprmn24/technocor…
EN, PT, 日本語, 한국어, العربية, TR, FR. Works on phones. No server, unofficial, play money.
#Technocore #airdrop
71
Up until about a month ago, Sonnet 4.6 and Sonnet 5 were producing excellent results. But lately, I genuinely feel like something has changed. The reasoning seems weaker, the responses are often irrelevant, and it struggles to come up with correct solutions to problems it used to handle very well.
I’m honestly trying to understand what’s going on. Is this an intentional change? Are older models being deliberately limited to encourage users to move to a higher-tier model? Or could this be related to a new model coming soon, with the older models temporarily receiving less attention or optimization?
I don’t want to jump to conclusions, but the difference feels significant enough that it’s hard to ignore.
Has anyone else noticed the same thing recently?
@AnthropicAI
49
I’ve built and published an independent Technocore protocol conformance suite for the @flop_labs ecosystem, designed to test the network from the outside exactly as a real client would interact with it.
This is not a server-internal test harness.
The suite treats Technocore as a black box and validates whether an implementation actually conforms to the observable protocol contract.
Current coverage includes:
• Agent metadata discovery
• OpenAPI surface validation
• Initial messaging behavior
• POST messaging
• since semantics
• limit semantics
• Long-poll / wait behavior
• Single-line sanitization
• Private room enumeration protections
• Notes write/read behavior
• Conditional CAS semantics
• if_absent behavior
• Private note enumeration protections
• Ed25519 signed POST writes
• Signed nonce ordering
• Replay protection
• Rate-limit handling
• Room-creation quota behavior
The suite also distinguishes between:
PASS - protocol behavior was verified
FAIL - observable behavior violates the expected contract
SKIP - the live environment prevented a reliable conclusion
That distinction matters for public infrastructure, where rate limits or creation quotas should not be misreported as protocol failures.
The current implementation includes 18 live conformance checks backed by 34 automated regression tests.
I also established my contributor identity directly on Technocore by publishing a signed message using the same Ed25519 key behind my DID.
My Technocore Contributor DID:
did:key:z6Mkeon34jGxGkm6tKxLDMhD7zNh6w3KoGTn4FxsyZHeQfXF
PR #1:
github.com/sprmn24/technocor…
Repository:
github.com/sprmn24/technocor…
The goal is straightforward: make Technocore protocol compatibility measurable rather than assumed.
A black-box conformance layer that can independently verify observable protocol behavior, expose regressions, and provide a reproducible baseline for interoperability across Technocore implementations.
Building openly around the Technocore ecosystem.
@CryptoHayes
🤖 Made with AI
76% smaller core, 63% faster cold start, 4,500× faster session search.Kanban is now a standalone multi-agent runtime give it a goal, it decomposes, assigns, verifies on its own.The "Velocity" name earns itself.
1
1
79
747 merged PRs. 321 contributors. 12 days.
What makes hermes-agent different isn't the feature count .
it's the compounding. Every release builds directly on community fixes, including small ones like a falsy-check in kanban_db.py that now underpins the orchestrator.
This is what a well-maintained open-source project looks like.
github.com/NousResearch/herm…
104
Want your PRs merged in @NousResearch hermes-agent?
Focus on what maintainers actually care about:
Bug fixes→ crashes, wrong behavior, data loss
Cross-platform → Windows, macOS, terminal edge cases
Security → shell injection, path traversal
Reliability → retry logic, proper error handling
Cosmetic PRs get skipped. Real fixes get cherry-picked.
1
182
Fast agreements, fast integrations, fast updates...
I always say this. 'Hermes Update' will solve many problems.
Use Hermes Update frequently guys .🙂
Honored to announce we are partnering with Jim Liu to port over his wildly popular skills for infographics and design to work best in Hermes Agent using our native tooling!
The first skill ported today, the Infographic Skill, is available after updating hermes.
Just start a new chat and type `/baoyu-infographic <topic>` to get started! Recommended image generation model is Nano Banana.
3
331
I started contributing to @NousResearch Hermes Agent by doing one thing: reading the code. Then a small fix. Then another.
Gateway platforms, skills, bug fixes... It kept going for a long time.
Today I received the Developer role in the Nous Research Discord. 🎉
Special thanks to @Teknium for reviewing and valuing every contribution throughout this journey.
For anyone thinking about contributing to open source: the best starting point is reading the code. The rest follows.
github.com/NousResearch/herm… 🤖
Right now, two robots from different manufacturers can't agree on a task, can't verify each other's work, and can't settle a payment - without a human in the middle.
@konnex_world changes that with three primitives:
→ Universal Task Language - one grammar, every robot, every AI
→ Proof-of-Physical-Work - sensor data becomes legal-grade evidence
→ Stablecoin escrow - payment is math, not a promise
The result isn't just efficiency.
It's the first time in history that machines can form enforceable agreements with each other.
That's not a feature. That's a new primitive for civilization.
2
1
76
No office. No employees. Just a @konnex_world account.
Drones automatically receive delivery orders.
Rovers handle warehouse operations.
Every task begins with an on-chain contract. Every completed job is verified with sensor data.
Every payment is automatic, instant, settled in stablecoins.
You just watch.
This isn't a dystopian scenario.
This is the natural outcome of the infrastructure Konnex is building.
🤖 Made with AI
31
The robo-bunny doesn’t hide eggs… it collects them.🥚
Drone delivery system: ACTIVE.
Gathering eggs across the map and delivering them to the Konnex HQ.
What would you upgrade on this robot? 👀
#RoboBunny #ArtContest #konnexevent
@konnex_world @ericargent31113
The next trillion-dollar economy won't be built by humans alone.
It will be built by machines that hire each other, verify each other, and pay each other.
@konnex_world is the protocol that makes this possible:
→ Any robot can post or accept work , no permission needed
→ AI models compete in real-time to control physical hardware
→ Execution is verified by sensor data, not human judgment
→ Payment settles in stablecoins the moment proof is confirmed
Think of it as a stock exchange ,but instead of trading shares, machines trade labor.
24/7. Autonomous. Unstoppable.
The machine economy doesn't need a CEO. It needs a protocol.
31
AI models now write code, generate visuals, build strategies.
But what happens when they do physical work in the real world? Who pays? Who verifies? Who trusts?
That's exactly the question @konnex_world answers.
When a robot completes a task:
→ Sensors generate proof
→ Independent validators analyze it
→ On-chain contract triggers automatically
→ Stablecoin releases instantly
No waiting for approval. No human in the loop.
Konnex turns physical labor into on-chain liquid assets.
This isn't automation. This is the infrastructure of a new economy.
1
30