Solana Agent Kit vs GOAT vs ElizaOS: which framework should you use?

Three frameworks come up in every Solana AI agent guide: Solana Agent Kit, GOAT, and ElizaOS. All open source. All built around plugins. All promising to let your agent move tokens. Pick the wrong one and you spend week three ripping out the foundation when you realize it does not fit the build you want to ship.

This guide breaks down what each framework is, who is shipping production agents on it, where it falls short, and how to choose. The framework decides what your agent can do. The infrastructure decides whether it works in production. If you want the broader build flow once you have picked, the step-by-step Solana AI agent build guide covers wallet setup, RPC integration, and Jupiter swaps end-to-end.

What is Solana Agent Kit?

Solana Agent Kit by SendAI is an open-source TypeScript toolkit that gives any LLM agent direct access to Solana protocols. The pitch on the project homepage is simple: "any agent, using any model can autonomously perform 60+ Solana actions." It is Apache-2.0 licensed and the de facto reference implementation for Solana-native agents. Production users include SEND Arcade's onchain games (via the Blinks plugin), 3land NFT minting flows, and SendAI's own Solana MCP server, which exposes the kit's actions to Claude Desktop and Cursor.

Architecturally it is a tool library, not an agent runtime. The core SolanaAgentKit class wraps a wallet plus an RPC connection, and you call .use(...) to register plugins. The kit ships five first-party plugins covering tokens, NFTs, DeFi, miscellaneous data, and Solana Blinks. You bring your own agent loop, usually LangChain or LangGraph or Vercel AI SDK, via the included adapter helpers.

Where it shines: depth on Solana. Out of the box the kit drives the protocols any production Solana agent ends up touching: DEXs and aggregators (Jupiter, Raydium, Orca, Meteora DLMM), price feeds (Pyth, Switchboard), MEV protection (Jito Bundles), perps (Drift), memecoin launches (Pump), liquid staking (Marinade, Sanctum), bridges (Wormhole, deBridge), ZK compression (Light Protocol). The v2 release notes show the catalog growing through 2025 with embedded wallet support via Turnkey, Privy, and Phantom.

Two providers ship as first-class plugins inside @solana-agent-kit/plugin-misc: Helius and Alchemy. Both expose Solana RPC, priority fees, webhooks, and curated data reads as native agent actions, callable from any agent loop with no glue code in between. They differ in reach: Helius is Solana-only; we add EVM chains (Ethereum, Base, Arbitrum, Polygon, and more) on top of Solana. That matters if your agent ever calls EVM contracts or follows a user's footprint across ecosystems.

Where it falls short: Solana only at the framework level. There is no native EVM agent surface, no first-class Eclipse or other SVM coverage, and no built-in agent runtime, memory, or social platform clients. The Python port (solana-agent-kit-py) has no formal releases and trails the TypeScript version on every plugin. Tagged release cadence is also uneven: the latest tagged release on GitHub is v2.0.9 from July 24, 2025, though the v2 branch keeps merging substantive contributions.

Pick it when: you are Solana-only, you want first-class Helius or Alchemy infra baked in, and you control the agent loop.

What is GOAT?

GOAT SDK by Crossmint (short for Great Onchain Agent Toolkit) is a framework-agnostic, multi-chain agent SDK. The original launch announcement framed it as "a single, unified library of onchain actions, ready to be used in 5 agent frameworks, two programming languages, on any wallet architecture across 30+ chains." It is MIT licensed.

The architecture is a clean three-axis matrix: framework adapter times wallet times plugin. You bring a wallet (viem, Solana web3.js, Crossmint smart wallet, Lit, Safe, and others), wrap it with the matching @goat-sdk/wallet-* package, attach plugins, and pass the result to your framework's tool interface. The same plugin works across whatever wallets the underlying chain supports.

Wallet packages exist for EVM, Solana, Aptos, Chromia, Cosmos, Fuel, Lit, Radix, Safe, Starknet, Sui, Zilliqa, MultiversX, plus Crossmint smart wallets in EVM and Solana flavors. Plugins span DEX swaps (Uniswap, Jupiter, 0x, Orca, Balancer), lending (Aave, Lulo, Renzo), prediction markets (Polymarket), bridging (deBridge, Mayan), NFTs (OpenSea, Magic Eden, Tensor), data (CoinGecko, Dexscreener, BirdEye, Nansen, Allora), and commerce rails through Crossmint. ElizaOS officially treats GOAT as the canonical way to add onchain abilities to a character via the ElizaOS GOAT plugin, and LangChain's tools documentation features it as a first-class integration.

Development has stalled. The main branch on the GOAT repo has not been touched since August 19, 2025, and the latest @goat-sdk/core release on npm is 0.5.0 from May 13, 2025. Solana coverage is also shallower than Solana Agent Kit's: GOAT exposes Jupiter, Orca, Tensor, and Magic Eden but skips the long tail of Solana-native protocols (Pump memecoin launches, Meteora DLMM, Drift perps, Marinade, Sanctum, Light Protocol).

Pick it when: you are building multi-chain (especially EVM-plus-Solana) and value a clean wallet-and-plugin abstraction over Solana-native depth. If you go this route, plan to maintain a fork.

What is ElizaOS?

ElizaOS by Eliza Labs is a different category of tool. Solana Agent Kit and GOAT are SDKs you import into your own agent code. ElizaOS is a complete agent runtime: it owns the message loop, memory, model routing, social client integrations, and multi-agent orchestration. You bring a character file. It brings everything else.

Launched in October 2024 as ai16z by Shaw Walters, the project rebranded to elizaOS in late January 2025 and moved under Eliza Labs. As of May 2026 it has 18,400+ stars on GitHub, MIT licensing, and v2 GA shipped in May 2026 (v2.0.3 latest, with three tagged releases in 36 hours during launch week). ElizaOS is the most actively shipped of the three frameworks.

Agents are defined by character files (JSON or TypeScript) that declare personality, plugins, and model settings. A minimal example from the ElizaOS character interface docs:

The runtime is model-agnostic (OpenAI, Anthropic, Gemini, Llama, Grok, local models) and ships first-party clients for Discord, X/Twitter, Telegram, Farcaster, Slack, WhatsApp, and Bluesky. RAG and persistent memory are built in. Blockchain support arrives through plugins: @elizaos/plugin-solana covers Solana token ops directly, while @elizaos/plugin-goat and plugin-solana-agent-kit let an Eliza character borrow GOAT's multi-chain breadth or Solana Agent Kit's Solana depth without rewriting the runtime.

Where it falls short: ElizaOS is heavier to set up than dropping an SDK into existing agent code. It is TypeScript-first with no first-class Python or Rust path. And because chains live behind plugins rather than at the core, raw Solana work is more direct in Solana Agent Kit.

Pick it when: you want a deployable autonomous agent with a persona, memory, and social platform clients out of the box.

How do these frameworks compare?

How should you choose?

This decision matrix maps common build profiles to frameworks.

Three nuances on top of that matrix.

These frameworks compose. ElizaOS can mount Solana Agent Kit (@elizaos/plugin-solana-agent-kit) or GOAT (@elizaos/plugin-goat) as plugins, so an SDK choice now does not lock you out of a runtime later. Start with what matches your chain scope and layer the runtime in when you need it.

Active maintenance is uneven. ElizaOS shipped v2 GA in May 2026 (v2.0.3 latest, with three patch releases in 36 hours during launch week). Solana Agent Kit's tagged release is v2.0.9 from July 2025 but the v2 branch is still merging substantive contributions. GOAT has not had a commit since August 2025.

Star counts are a proxy, not proof. ElizaOS pulls 18,400+ because it is a full runtime with broader appeal. Solana Agent Kit and GOAT serve narrower audiences by design, not lower quality.

What's the infrastructure layer underneath?

All three frameworks send transactions through an RPC provider, and that provider determines whether your agent ships fast and stays up. Reads need archival depth and gRPC streaming. Writes need fast, reliable transaction landing. Multi-region coverage matters when your agent runs in production for hours at a time.

If you are using Solana Agent Kit, we are no longer just the infrastructure layer underneath. We ship as a first-class plugin inside @solana-agent-kit/plugin-misc, exposing Solana RPC, priority fee estimation, token prices (by symbol, by address, historical), multi-chain portfolio reads, Notify webhooks, and an endpoint helper as native agent tools.

The hosted Solana agent scaffolder wires both providers into every new project, so a fresh scaffold ships with the RPC choice baked in. We're now a first-class option in the skills catalog and the RPC provider guide. Spin up an agent with our plugin pre-wired at solana.new.

Beyond the kit, the Alchemy CLI lets agents sign in with Solana (SIWS), pay through x402 in USDC, and call Core RPC, Token, Portfolio, NFT, and Prices APIs across 100+ chains. No API key, no dashboard signup, no human in the loop. Start building.

If you are still shopping for the build flow itself, the Solana AI agent build guide walks through framework selection, RPC setup, wallet security, and a working Jupiter swap example. Pick the framework that matches your chain scope and runtime needs, then wire it to the infrastructure underneath. The framework decides what your agent can do. The infrastructure decides whether it works in production.