"Trading agent framework" is one of the questions builders keep bringing to Sato Hub. Not a trend take — a build question. So here's the honest answer: what's actually tracked, live, and checkable, laid out by the layer it fills. No vibes, no "soon." Just what's shipped, sourced to its own [Sato Hub](https://satohub.ai/directory) listing.
A trading agent needs three things to exist at all: something to orchestrate the decision loop, something to move funds onchain, and something to tell it whether the thing it's about to trade is worth the risk. Here's what fills each layer today.
Layer 1: the orchestrator
This is the part that runs the actual decision loop — research, debate, risk sign-off, or some version of it.
- ▸[TradingAgents](https://satohub.ai/resources/tradingagents) — an open-source multi-agent LLM framework that simulates a trading firm: analyst, trader, and risk-management agents arguing before a decision. Built for research, not production execution.
- ▸[ElizaOS](https://satohub.ai/resources/elizaos) — a general-purpose open-source TypeScript framework for crypto-native multi-agent systems, running across Solana, Ethereum, and Base. Not trading-specific, but the base plenty of trading bots get built on.
- ▸[Olas (Autonolas)](https://satohub.ai/resources/olas) — a network and framework for co-owned autonomous agent services that operate onchain, spanning Ethereum, Gnosis, and Base.
- ▸[Almanak](https://satohub.ai/resources/almanak) — the one built specifically for this: an agent framework paired with a non-custodial vault protocol for designing, backtesting, and deploying automated DeFi yield and trading strategies, across eleven tracked chains.
Four options, two philosophies: general multi-agent frameworks you adapt (ElizaOS, Olas) versus purpose-built trading-and-vault stacks (TradingAgents for research, Almanak for backtested deployment).
Layer 2: wallets and execution
An orchestrator that can't move funds is a chatbot. This layer is what turns a decision into a transaction.
- ▸[Coinbase AgentKit](https://satohub.ai/resources/coinbase-agentkit) — Coinbase's toolkit for giving agents wallets and the ability to take onchain actions, covering Base, Ethereum, and multichain deployments.
- ▸[GOAT SDK](https://satohub.ai/resources/goat-sdk) — Crossmint's open-source library of onchain "actions" (tools) for agents, spanning Ethereum, Base, Solana, and beyond.
- ▸[0xGasless AgentKit](https://satohub.ai/resources/0xgasless-agentkit) — a toolkit for gasless wallet access on BNB Chain, Avalanche, and Base, aimed at the friction point of making an agent pay its own gas.
Three different answers to "how does the agent hold and spend a wallet," none of them a standard yet. Which one you pick mostly comes down to which chains and which gas model you're targeting.
Layer 3: risk and data
Before a trading agent commits capital, something needs to score what it's about to touch.
- ▸[Philidor MCP](https://satohub.ai/resources/philidor-mcp) — a hosted MCP server scoring risk across 700+ DeFi vaults on Morpho, Aave, Spark, Yearn, and Beefy, built for exactly this kind of agent due diligence.
This is the thinnest layer of the four. There's one tracked, purpose-built risk-scoring MCP for vault due diligence — general market-data or price-feed skills exist elsewhere in the directory, but a dedicated pre-trade risk layer for agents is still mostly a gap.
What the Sato Score does and doesn't tell you here
All eight components above carry a High Sato Score (76–88). That's a transparency and liveness signal — active repos, real docs, verifiable code, a working deploy spec — not a safety, quality, or performance grade. A framework scoring 88 tells you the project is well-documented and maintained. It tells you nothing about whether a strategy built on it will make or lose money, because none of these have published, verified trading results. Treat the score as evidence worth investigating, not evidence worth trusting with capital.
The honest gap
Line up the layers and the hole is obvious: orchestration and wallet execution are crowded (four frameworks, three wallet kits), but risk scoring — the layer that actually protects capital — has one tracked entry. Backtesting, strategy verification, and post-trade attribution tools aren't represented here at all. If you're building in this space, that's where there's the most room, and the least competition.
What to watch
Watch whether any of these frameworks converge on a shared wallet-execution interface — right now every trading agent stack picks its own, which means no portability between them. And watch the risk layer specifically: one MCP server covering 700 vaults is a start, not coverage.
Out of 383 tracked listings on Sato Hub, this is the current shape of the trading-agent base. If you're building one, the stack above is what exists to build it from — map your own at [satohub.ai/build](https://satohub.ai/build).
Sources
- ▸[TradingAgents — Sato Hub](https://satohub.ai/resources/tradingagents)
- ▸[ElizaOS — Sato Hub](https://satohub.ai/resources/elizaos)
- ▸[Olas (Autonolas) — Sato Hub](https://satohub.ai/resources/olas)
- ▸[Almanak — Sato Hub](https://satohub.ai/resources/almanak)
- ▸[Coinbase AgentKit — Sato Hub](https://satohub.ai/resources/coinbase-agentkit)
- ▸[GOAT SDK — Sato Hub](https://satohub.ai/resources/goat-sdk)
- ▸[0xGasless AgentKit — Sato Hub](https://satohub.ai/resources/0xgasless-agentkit)
- ▸[Philidor MCP — Sato Hub](https://satohub.ai/resources/philidor-mcp)
- ▸[Sato Hub directory](https://satohub.ai/directory)