Build async multi-agent orchestration apps in Python with the agent-squad framework.
npx skills add https://github.com/2FastLabs/agent-squad --skill agent-squad-python
Please help me install this Agent Skill. Skill: agent-squad-python Source: https://github.com/2FastLabs/agent-squad/tree/main/python Command: npx skills add https://github.com/2FastLabs/agent-squad --skill agent-squad-python
Building applications that route user queries across multiple specialized AI agents requires wiring together classifiers, agents, storage, and tools correctly. This Skill gives an AI assistant the mental model, real API signatures, task recipes, and gotchas needed to write correct agent-squad Python code on the first attempt.
Ask your assistant to read python/SKILL.md and then write an agent-squad Python app with two BedrockLLMAgent agents routed by a BedrockClassifier.
High-intent search queries and answers about installing and using this skill.
Create an AgentSquad orchestrator, register agents with add_agent, and await route_request with user_input, user_id, and session_id. A classifier such as BedrockClassifier selects the best agent based on each agent's name and description.
Install optional extras with pip: agent-squad[anthropic] for AnthropicAgent and AnthropicClassifier, agent-squad[openai] for OpenAI equivalents, or agent-squad[all] for everything except strands-agents. The core package alone includes no LLM runtime.
Yes. Agents like BedrockLLMAgent support streaming, and you must pass stream_response=True to route_request. Then response.output is an async generator of AgentStreamResponse chunks; always branch on response.streaming before consuming output.
The Python orchestrator requires a classifier at construction time. It defaults to BedrockClassifier only if boto3 is installed; otherwise you must pass a classifier explicitly, such as AnthropicClassifier or OpenAIClassifier.
Install agent-squad[mcp], then await MCPToolProvider.create with stdio or SSE server configs and pass the provider as the agent's tool_config. Call await provider.disconnect() when finished to close server connections.
Use GroundedAgent when answers must not drift from tool data, such as prices or balances. A gatherer LLM calls tools while an isolated presenter LLM writes the reply only from curated facts, never seeing chat history.
BerriAIRegister a new LLM deployment in LiteLLM via POST /model/new and validate routing.
aussierobotsDeploy MCP servers on AWS Lambda with DynamoDB session storage and SSE streaming.
dkyazzentwatwaCompare text similarity using TF-IDF, cosine similarity, and Jaccard index.
2FastLabsOrchestrate AI agents for iOS 16+ and macOS 14+ apps.
