Dagent

We built a platform that makes it easy for developers to use the right AI agent without the hassle of finding, hosting, or paying them manually.

Here’s how it works: when a developer creates an API key, they stake some tokens. The first time they send a request, our system automatically finds the best agent that matches their needs — things like cost, skills, or provider. That agent gets assigned, and from then on all their requests go straight through it. We keep track of usage and deduct the cost from their staked balance until it runs out.

For developers, it feels like using any other API: no marketplaces to browse, no payment headaches, just call the endpoint and get results. For agent creators, it’s a way to deploy their agents publicly and get paid automatically whenever they’re matched.

We’ve also built in reliability — if an agent goes offline or fails, our system can quickly rematch to another one without breaking the developer’s workflow. And because it plugs directly into Google’s ADK as a Function Tool, agents can use it natively inside multi-agent setups.

In short: it’s a simple, fair, and automatic way to connect people who need AI agents with people who build them — powered by staking and pay-per-use.

This project is built around a staking smart contract that makes monetization for AI agents possible. Developers stake tokens to generate API keys, and each request they make reduces their balance based on the matched agent’s cost and token usage.

For the agent matchup process, we explored multiple approaches and ultimately settled on a hybrid method: semantic search over agent descriptions combined with requirement-specific field weightage (like preferred LLM provider, cost limits, and skills). This gives us more accurate and fair matches.

On the client side, we built a Python package with adapters for different agentic frameworks, so our tool is framework-agnostic. This allows it to plug into ecosystems like Google’s ADK seamlessly — the tool looks just like any other FunctionTool.

For the infrastructure, we implemented a proxy layer that handles request routing, sets an agent_id cookie for session persistence, and ensures transparent fallback or rotation if an agent goes offline. This design keeps the experience smooth for developers while maintaining reliability.

One of the “hackier” parts we’re proud of was building the cookie-based agent persistence on top of blockchain-backed staking — it let us simulate a Web2-simple developer experience on top of a Web3 settlement layer.