Services - AI agents that act on-chain, within limits you set
An AI agent is software that uses a language model to decide what to do next, then does it through tools: reading market data, calling APIs, signing transactions. We build agents for crypto products, from trading and DeFi operations to agents that pay for the data they use in USDC, with the limits and audit trail that money requires.
We publish how we build them: a custom ElizaOS agent from setup to deployment, an autonomous trading agent that executes on Hyperliquid from a trusted execution environment, and the x402 and Circle Nanopayments integrations that let an agent pay per request.
What we build - Agents for crypto products
An agent is only as useful as the tools it can call and as safe as the limits around them. These are the agents we build.
- Trading agents. Agents that combine market data, sentiment, and strategy rules to decide, then execute on exchanges such as Hyperliquid, with position limits enforced in code rather than in the prompt.
- DeFi operations agents. Agents that watch liquidity positions, rebalance, claim rewards, and report, built on the same stack as our liquidity management automation.
- Agents that pay. Agents that buy API calls, data, and compute per request in USDC through x402 and Circle Nanopayments, with a spending cap per task.
- Community and support agents. ElizaOS agents on Telegram, Discord, and X that answer questions about your protocol, onboard users, and hand off to a human when they should.
- Custom plugins and tools. ElizaOS plugins and tool integrations that connect an agent to your contracts, your backend, and third-party APIs.
- APIs that agents can pay for. The other side of the market: we put your API behind an x402 paywall, so AI agents can pay for it per request without accounts or API keys.
How we work - Guardrails first, intelligence second
An agent that can sign transactions is a wallet that makes its own decisions. We build its limits before we build its reasoning.
- Hard limits in code. Spending caps, allowlisted contracts, maximum position sizes, and a kill switch live outside the model, so no prompt can override them.
- Keys in a secure enclave. Agents that hold keys run in a trusted execution environment, as in our Gramine deployment on Azure, or sign through narrowly scoped keys.
- The model proposes, code executes. Typed, tested code validates every action the model suggests before it runs, and logs the reasoning that led to it.
- Evaluated before launch. We replay historical data and adversarial inputs against the agent and review its decisions before it touches real funds.
- Observable in production. Dashboards and alerts for every decision, trade, and payment, with a human approval step above a threshold you choose.
- Model-agnostic. ElizaOS and TypeScript with the model of your choice behind them, so you can switch providers without a rewrite.
Pricing - What AI agent development costs
We publish our ranges. At a blended rate of $500 to $700 per engineer-day, project management, QA, and deployment included:
- An AI agent with its backend and tool integrations: $12,000 to $29,000, in 3-5 weeks.
- Scope grows with the tools, wallets, and chains the agent touches. The calculator itemizes each add-on.
- Running costs (model API usage and hosting) are separate from the build price.
From the blog - How we build agents, with the code
Step-by-step guides from our own agent work.