The question that keeps coming up
"What's the trading agent framework?" is a search we keep seeing, and the honest answer is: there isn't one. There's a stack — an orchestration layer that decides what to do, an action layer that executes it on-chain, and a skill layer that speaks each chain's native language. Nobody ships all three under one name yet. Here's what's actually tracked and buildable across each, no performance claims attached — a Sato Score reflects transparency and activity, not returns.
Layer 1: multi-agent orchestration
This is the "brain" layer — deciding what to trade and when, usually by running multiple LLM agents against each other.
- ▸[TradingAgents](https://satohub.ai/resources/tradingagents) simulates a trading firm outright: analyst, trader, and risk-management agents debating a position before it's taken. It's built for research, not unattended capital.
- ▸[ElizaOS](https://satohub.ai/resources/elizaos) is a general-purpose multi-agent framework in TypeScript that a lot of crypto-native agents are built on top of — trading is one use case among several.
- ▸[Olas (Autonolas)](https://satohub.ai/resources/olas) takes a different angle: co-owned autonomous services running on-chain, so the agent's ownership and incentives are structured onchain rather than left as a backend implementation detail.
- ▸[Almanak](https://satohub.ai/resources/almanak) pairs an agent framework with a non-custodial vault protocol built specifically for designing, backtesting, and deploying DeFi yield and trading strategies — the closest thing on this list to a purpose-built trading stack, spanning eleven chains.
None of this is a signal generator you point at a market and walk away from. It's the reasoning scaffold — you still decide the strategy logic.
Layer 2: wallets and action toolkits
Once an agent decides to act, it needs hands: a wallet it can control, and a way to turn "swap this" into a signed transaction.
- ▸[Coinbase AgentKit](https://satohub.ai/resources/coinbase-agentkit) gives agents wallets and onchain action capability across Base, Ethereum, and other EVM chains — Coinbase's answer to "how does my agent hold and move funds."
- ▸[GOAT SDK](https://satohub.ai/resources/goat-sdk), from Crossmint, is an open-source library of onchain actions ready to plug into an agent across Ethereum, Base, and Solana — the connective tissue between a framework's decision and a chain's mempool.
- ▸[0xGasless AgentKit](https://satohub.ai/resources/0xgasless-agentkit) solves a specific, unglamorous problem: gas. It gives agents gasless wallet access on BNB Chain, Avalanche, and Base, which matters if your agent is supposed to trade without a human topping up ETH for fees.
Layer 3: chain-specific skill repos
Frameworks need chain-specific verbs. This layer is narrower and less abstracted, trading generality for depth on one ecosystem.
- ▸[Solana Agent Kit](https://satohub.ai/resources/solana-agent-kit), from SendAI, connects agents directly to Solana protocol actions — the toolkit of choice if your trading agent lives entirely on one chain rather than routing across several.
What's missing
Looking at the eight components tracked here, the gaps are as informative as the coverage. There's no single framework that owns orchestration, wallet custody, and execution end to end — you're integrating at least two of these layers yourself. Backtesting is thin outside Almanak's vault protocol. And every one of these is self-described infrastructure, not a strategy — none of them is claiming, and none should be read as claiming, that the agent it powers makes money. All eight currently track a High Sato Score, meaning they're active, documented, and their code is inspectable — that's a transparency signal, not a returns guarantee, and it's worth checking directly rather than taking on faith.
What to watch
The next real consolidation move is whichever framework ships native wallet custody instead of bolting on AgentKit or GOAT SDK — that's the seam every team currently has to wire by hand. Watch for backtesting tooling to catch up to the orchestration layer, since right now most of these frameworks are stronger at deciding than at proving a decision would have worked.
Build it
If you're assembling a trading agent today, you're picking from these layers, not off a shelf. [Sato Hub's builder](https://satohub.ai/build) maps the stack for you — describe the agent, and it surfaces the framework, wallet, and skill components that fit, scored and sourced.