Create

Creating an Agent

mesh create my-agent \
  --skill external-agent-bridge \
  --skill fast-analytics-query \
  -d "My agent description" \
  --model claude-sonnet-4-6

This generates:

agents/my_agent/
  agent.yaml       # Manifest
  main.py          # Bootstrap — registers with control plane
  skills.py        # @router.skill() endpoints for each tool
  reasoners.py     # @router.reasoner() with system prompt + tools="discover"
  skill_*.py       # Tool implementations copied from library

What Gets Generated

File	Purpose
agent.yaml	Manifest — skills, model, version, status
main.py	Agent bootstrap — connects to control plane
reasoners.py	LLM reasoning loop with system prompt
skills.py	Tool endpoints discovered at runtime

Via the Builder

In the TUI or Web UI, ask the builder:

“Create an agent that queries Fetch analytics data”

The builder will select appropriate skills, generate a system prompt, and scaffold the agent.

Models

Any model from Bedrock, OpenAI, or Anthropic. The model string format is provider/model-id — LiteLLM handles routing.

Bedrock (AWS)

Model	Tier	Notes
claude-sonnet-4-6	Premium	Default. Best quality, clean tool calls
claude-haiku-4-5	Fast	Best value, very fast
claude-opus-4-6	Premium	Most capable, deep reasoning
llama-4-scout-17b	Budget	Clean tool calls, great price/performance
llama-4-maverick-17b	Budget	1M context, needs tool call dedup
minimax-m2-1	Reasoning	Needs higher maxTokens

OpenAI

Model	Tier	Notes
gpt-5-mini	Fast	Reliable tool use
gpt-4.1-mini	Fast	Good balance of speed and quality
o3-mini	Reasoning	OpenAI reasoning model

Anthropic (direct API)

Model	Tier	Notes
claude-sonnet-4-6-direct	Premium	Via Anthropic API, needs ANTHROPIC_API_KEY

Or pass any LiteLLM-compatible model string directly: bedrock/us.anthropic.claude-sonnet-4-6, openai/gpt-5-mini, etc.