An autonomous trading agent is not one product. It is a stack: something that runs the agent, something that talks to a venue, something that turns an intent into a transaction, and something that holds funds and pays for inputs. The prompt is the cute part. The stack is where the work lives.
Here is the honest state of that stack, built only from the listings Sato Hub tracks. Everything below is what each project *says* it does, as described on its Sato Hub page. A Sato Score is a transparency and liveness signal. It is not a safety, quality or returns grade, and nothing here is a claim about how any of these perform.
Layer 1: the runtime and framework
This layer decides how an agent is structured, who owns it and where it runs.
- ▸Olas (Autonolas) describes itself as a network and framework for co-owned autonomous agent services operating onchain. Tracked chains: Ethereum, Gnosis, Base and multichain. Sato Score 87 (High), with a deploy spec.
- ▸Injective Agents is a platform for deploying autonomous AI trading agents on Injective, with onchain identity, order-book trading and MCP framework support. Sato Score 58 (Medium), with a deploy spec.
The difference matters. Olas is a general framework for agent services. Injective Agents is a trading-first platform tied to one chain's order book. Pick the framework for the shape of the agent you want, not for the logo.
Layer 2: the venue
An agent that trades needs a way to talk to where trading happens.
- ▸Hyperliquid Python SDK is the official open-source Python SDK for programmatic trading on the Hyperliquid perpetuals DEX. Sato Score 76 (High).
- ▸CloddsBot is an open-source autonomous AI trading agent that says it operates across 1000+ markets, including Polymarket, Kalshi and Hyperliquid, across Solana, Base, Ethereum, Arbitrum, Optimism, Polygon and Hyperliquid. Sato Score 83 (High).
One is a thin, official client for a single venue. The other is a whole agent that reaches across many. Notice that CloddsBot's market reach is its own description. A listing page can tell you what a project claims; it cannot tell you how the agent behaves with real orders. That is a check you run yourself.
Layer 3: execution
Between a decision and a settled transaction sits the part most demos skip.
- ▸Enso is an intent-based onchain execution engine and API that lets developers and agents bundle multi-step DeFi actions into a single transaction. Ethereum, Base, Arbitrum, Optimism, Polygon and multichain. Sato Score 80 (High).
- ▸Ophis is an intent-based DEX aggregator for agents: natural-language swap intents settled via batch auction across 11+ EVM chains plus Solana. Sato Score 65 (Medium).
Both take an intent rather than a hand-built route. That is a real convenience for an agent, and also a real question: when something else decides the route, what does your agent log about why? Know what you are delegating.
Layer 4: wallet and payment rail
An agent that spends needs a wallet, and an agent that buys data or inference needs a way to pay for it.
- ▸Franklin is an autonomous AI agent with its own wallet that spends USDC to get real work done. Base and Solana. Sato Score 70 (High).
- ▸BlockRun is a pay-per-call gateway where agents reach 55+ LLMs, data and tools through one endpoint, settled in USDC via x402, with no API keys. Base and Solana. Sato Score 69 (Medium).
This is the layer that turns an agent from a script into an economic actor, and it is the youngest layer in the list.
What is still missing
Read the eight descriptions again and notice what none of them says.
- ▸Spend controls as a first-class piece. Wallets appear in the list. A clearly described limit on what a trading agent may move is not what these listings lead with. If your agent can move funds, "trust me" is not a control.
- ▸Evidence of behavior. The listings describe capabilities. None of the descriptions above is evidence of how an agent behaves over time. Self-reported is not the same as shown.
- ▸A shared way to compare. Chains, venues and intents differ across every entry. There is no common yardstick between a venue SDK and an execution engine, so you assemble and test the pieces yourself.
Those gaps are the opportunity. The components exist. The glue, the limits and the receipts are what a builder adds.
What to watch
Two things. First, whether execution engines and aggregators expose enough of their routing decisions for an agent to explain itself afterwards. Second, whether payment rails like x402 become the default way a trading agent pays for data. Both are visible in the listings over time, and both are cheap to check yourself.
Build it
You do not have to wire this by hand. Describe the agent and Sato maps the stack: framework, venue, execution and wallet, each with its page and deploy spec. Start at satohub.ai/build.