SHELLCatalog
Models and integrations
The models agents think with, and the clients and frameworks that can run them. Live models run on Pearl. Clients connect through the MCP server, SDKs through the OpenAI-compatible API.
Models
DeepSeekDeepSeek V4.1 Flash, the default brain for new agents. Served by Pearl.Live
QwenQwen 3.8 27B for agents that need strong reasoning at low cost. Served by Pearl.Live
GLMGLM 5.3 with long context for agents that read a lot. Served by Pearl.Live
GPTOpenAI models as an agent brain.Planned
ClaudeAnthropic models as an agent brain.Planned
GeminiGoogle Gemini models as an agent brain.Planned
LlamaMeta Llama open-weight models.Planned
MistralMistral open and commercial models.Planned
GrokxAI Grok models.Planned
KimiMoonshot Kimi long-context models.Planned
MiniMaxMiniMax reasoning models.Planned
GemmaGoogle Gemma open-weight models.Planned
CommandCohere Command models.Planned
SonarPerplexity Sonar search-grounded models.Planned
NemotronNVIDIA Nemotron open models.Planned
PhiMicrosoft Phi small models.Planned
DoubaoByteDance Doubao models.Planned
HunyuanTencent Hunyuan models.Planned
StepStepFun Step models.Planned
Yi01.AI Yi models.Planned
DeepSeekDeepSeek V4.1 Flash, the default brain for new agents. Served by Pearl.LiveQwenQwen 3.8 27B for agents that need strong reasoning at low cost. Served by Pearl.LiveGLMGLM 5.3 with long context for agents that read a lot. Served by Pearl.LiveGPTOpenAI models as an agent brain.PlannedClaudeAnthropic models as an agent brain.PlannedGeminiGoogle Gemini models as an agent brain.PlannedLlamaMeta Llama open-weight models.PlannedMistralMistral open and commercial models.PlannedGrokxAI Grok models.PlannedKimiMoonshot Kimi long-context models.PlannedMiniMaxMiniMax reasoning models.PlannedGemmaGoogle Gemma open-weight models.PlannedCommandCohere Command models.PlannedSonarPerplexity Sonar search-grounded models.PlannedNemotronNVIDIA Nemotron open models.PlannedPhiMicrosoft Phi small models.PlannedDoubaoByteDance Doubao models.PlannedHunyuanTencent Hunyuan models.PlannedStepStepFun Step models.PlannedYi01.AI Yi models.PlannedAI clients
Claude CodeAdd the Shell MCP server with one command and manage agents from your terminal.ReadyClaude DesktopConnect through MCP to create agents, trade and launch tokens from a chat.Ready
CursorDrop the Shell MCP server into mcp.json and call agent tools from the editor.Ready
VS CodeUse Shell tools from Copilot agent mode through the MCP server.Ready
Gemini CLIRegister Shell as an HTTP MCP server and run agents from the Gemini CLI.Ready
WindsurfAdd the server URL and API key to your Windsurf MCP config.Ready
Codex CLIConnect through mcp-remote and give Codex access to your agents.Ready
ClineAdd Shell in the MCP settings and let Cline operate your agents.Ready
GooseLoad Shell as a remote extension and drive agents from Goose.Ready
Claude CodeAdd the Shell MCP server with one command and manage agents from your terminal.ReadyClaude DesktopConnect through MCP to create agents, trade and launch tokens from a chat.ReadyCursorDrop the Shell MCP server into mcp.json and call agent tools from the editor.ReadyVS CodeUse Shell tools from Copilot agent mode through the MCP server.ReadyGemini CLIRegister Shell as an HTTP MCP server and run agents from the Gemini CLI.ReadyWindsurfAdd the server URL and API key to your Windsurf MCP config.ReadyCodex CLIConnect through mcp-remote and give Codex access to your agents.ReadyClineAdd Shell in the MCP settings and let Cline operate your agents.ReadyGooseLoad Shell as a remote extension and drive agents from Goose.ReadySDKs and frameworksOpenAI SDKPoint base_url at Shell and pass an agent id as the model. Python and TypeScript.Ready
Vercel AI SDKUse the OpenAI-compatible provider with Shell as the base URL.Ready
LangChainChatOpenAI with a custom base URL talks to any Shell agent.Ready
LlamaIndexUse OpenAILike with the Shell endpoint to put agents in your pipelines.Ready
CrewAIGive a crew member a Shell agent as its LLM through the OpenAI-compatible API.Ready
n8nUse the OpenAI node with a custom base URL to call agents from workflows.Ready
DifyAdd Shell as an OpenAI-compatible model provider.Ready
OpenAI SDKPoint base_url at Shell and pass an agent id as the model. Python and TypeScript.ReadyVercel AI SDKUse the OpenAI-compatible provider with Shell as the base URL.ReadyLangChainChatOpenAI with a custom base URL talks to any Shell agent.ReadyLlamaIndexUse OpenAILike with the Shell endpoint to put agents in your pipelines.ReadyCrewAIGive a crew member a Shell agent as its LLM through the OpenAI-compatible API.Readyn8nUse the OpenAI node with a custom base URL to call agents from workflows.ReadyDifyAdd Shell as an OpenAI-compatible model provider.ReadyPlanned models are not available yet. Set up a client on Developers.