Crewdegen → CrewMind.xyz

Building Self-Improving AI Agents for a Permissionless World

We started building on Solana back in 2021, during what many now call the Solana Summer. That early period shaped a lot of how we think today. We co-founded one of the first NFT-backed lending protocols on Solana — a project that was later acquired and merged into another team and continues to operate today as Banx.gg.

Since then, we've shipped multiple Solana products and hackathon builds. Some of them failed. All of them made us stronger — technically, architecturally, and as builders deeply embedded in the ecosystem. Along the way, we built Adrastea Finance, a liquid staking and restaking protocol, and kept pushing toward a larger question we couldn't ignore:

What happens when AI agents are finally allowed to act, not just advise?

What We're Building Today

Today, we're building CrewMind.xyz — agent systems that can execute real on-chain actions on Solana and continuously improve from data and feedback.

Our first live product is Crewdegen.com: a multi-agent AI autopilot for portfolio management. Right now, it operates in crypto — not because crypto is the end goal, but because it is the only environment where this kind of autonomy is already possible.

Crewdegen agents:

• read hundreds of news sources, social channels, and market signals,
• analyze sentiment and technical indicators,
• make trading decisions,
• execute real transactions with algorithmic risk management (TP/SL),
• and do all of this continuously.

Users configure the core parameters, while the agents handle execution. Since launching in September, we've onboarded our first users and collected a growing dataset of real decisions, outcomes, and market contexts.

That data changed everything.

Why We Believe Crypto Is Where AI Agents Will Truly Emerge

We strongly believe that Web3 — and crypto in particular — will be the environment where AI agents evolve fastest and most visibly.

Not because AI isn't useful elsewhere. Quite the opposite.

AI could already manage capital in banking, logistics, transportation, or healthcare — but it is legally constrained everywhere. You cannot give an AI agent direct control over a bank account. You cannot allow it to make autonomous decisions that affect people's lives without layers of regulation, approvals, and liability.

Even autonomous driving illustrates this perfectly: the technology works, but deployment is slow and restricted.

Crypto is different.

Permissionless protocols don't require approval to interact. Pseudonymous wallets can already be controlled by software. Smart contracts don't ask who you are — only whether the transaction is valid.

AI agents can already act here, today.

That's why we believe decentralized finance is the natural birthplace of truly autonomous agents — agents that can reason, decide, execute, and improve.

And we believe Solana is the best place to build this future:

• fast and cheap transactions,
• a massive, real builder community,
• production-grade infrastructure,
• and people who genuinely enjoy working together.

We're also proud members of Solana Superteam Kazakhstan, which has been an important part of our journey and network.

From Crewdegen to CrewMind

Crewdegen taught us two important things:

1. There is real demand for autonomous AI systems that act, not just analyze.
2. Onboarding and customization are hard — especially when users want control over logic, not just parameters.

That's why we're pivoting toward CrewMind.

CrewMind is not just about trading. Trading is our proving ground.

The Pro Version: Your Own AI Autopilot

In the Pro version of CrewMind, users will be able to launch their own autonomous agent.

At a high level, the flow looks like this:

1. Start with an idea — A user comes with a natural-language idea of how an agent should behave: risk preferences, market beliefs, reactions to news, position sizing logic.
2. Create a modular prompt — The system helps structure this idea into a modular prompt: separate reasoning blocks, decision rules, constraints, and objectives.
3. Choose execution parameters — Users select: the underlying LLM, execution frequency, supported assets, and safety limits.
4. Run a backtest on real historical data — The agent is replayed on historical market conditions: news, sentiment, prices, volatility — step by step.
5. Classify outcomes — The system automatically classifies decisions as successful or unsuccessful based on objective results.
6. Improve the prompt using feedback — Using these classifications, the system runs an optimization loop: it analyzes patterns, identifies weaknesses, and suggests concrete prompt changes — not vague advice, but actionable edits.
7. Test again and verify improvement — The improved prompt is backtested again, allowing users to see measurable improvement, not just theoretical suggestions.

This approach is inspired by modern agent-training frameworks (e.g. DSPy.ai) and allows users to move from intuition to evidence-backed strategies — all using natural language.

No other product today offers this level of transparency, iteration, and control over autonomous trading logic.

Beyond Trading: True On-Chain Agents

Trading is just the beginning.

Our long-term vision is to give agents real agency across DeFi:

• interacting with lending protocols,
• managing collateral and leverage,
• staking and restaking assets,
• responding to yield changes discovered via external data feeds,
• coordinating multiple protocols to execute complex strategies,
• and acting as a true, always-on portfolio manager.

Not a degenerate trader — but a disciplined, tireless investment agent.

One that never sleeps.
One that reads everything.
One that reacts faster than any human can.
One that operates entirely within the rules of open, permissionless systems.

Building the Future Together

We believe this will be:

• one of the first true autonomous agent platforms on Solana,
• one of the first in Web3,
• and a foundation for how people will interact with capital in the future.

If this vision resonates with you, we'd love to build it together.