Show HN: Praval Agentic AI Framework (github.com)

Praval is a new Pythonic multi-agent AI framework that makes it easy to build ecosystems of specialized, collaborating agents using simple decorator-based APIs. Instead of a single monolithic pipeline, developers define agents by identity (e.g., @agent("researcher")) and exchange structured JSON “spores” over a shared communication substrate called the Reef. Agents self-coordinate via message passing—no central orchestrator—so small functions chain into emergent workflows (authors claim examples shrink from ~489 to ~50 lines). The framework ships with a tool system, examples, and a familiar start_agents(...) runtime; requirements are Python 3.9+ and at least one LLM API key (OpenAI, Anthropic, or Cohere). Technically Praval targets production and research use cases: multi-layered persistent memory powered by ChromaDB (short-term, long-term, episodic, semantic), multi-LLM provider selection, multiple storage backends (Postgres, Redis, S3, Qdrant, local FS), and extensible transports (AMQP, MQTT, STOMP). It includes zero-config OpenTelemetry tracing, enterprise-grade encryption (Curve25519, XSalsa20+Poly1305, Ed25519) with automatic key rotation, and tool decorators for external capabilities. For the AI/ML community, Praval lowers the bar for composable, observable, and secure agentic systems—accelerating prototyping of emergent multi-agent workflows while providing production-ready primitives; caveats include dependency on external LLM APIs and some features (streaming, visual debugging) still in progress.