Show HN: Echos – A lightweight multi-agent AI system with pre-built agents (github.com)

Echos is an open-source, lightweight multi-agent AI platform that provides pre-built agent “services” (db_agent, api_agent, search_agent, data_agent, code_agent, orchestrator) you compose via a YAML workflow instead of rebuilding common infrastructure. Launched on GitHub with an MIT license, it aims to stop teams from reimplementing orchestrators, retry/backoff strategies, loop limiting, SQL/SSRF guardrails, cost tracking, and observability. The interface feels like composing AWS-style services: declare agents, routes and policies in workflow.yaml and run tasks locally or from code with the @echoshq/runtime or the npx CLI. Technically, Echos supports namespaced shared memory, per-agent maxLoops to prevent runaway behavior, automatic retries with exponential backoff, per-agent and per-workflow cost ceilings, and visual traces for real-time debugging and cost visibility. Security guardrails include SQL operation/table whitelisting (e.g., restrict to SELECT on users/orders) and SSRF/domain whitelisting; integrations support multiple LLM providers (OpenAI GPT family and Anthropic Claude), web search (Serper/Brave), embeddings, and code generation. It’s targeted at teams who want composable, auditable multi-agent workflows with built-in safety and observability without building an orchestration stack from scratch.