Skip to content
promptfleet Agents

SDK / Core overview

These crates build on each other:

  1. agent_core — protocol-agnostic domain types (messages, content parts, task phases, conversation context). Zero dependencies beyond serde.
  2. agent_sdk — runtime-first SDK facade: AgentBuilder, skills, services, LLM engine, A2A/AG-UI adapters, host composition. 25+ feature flags for granular opt-in.
  3. pf_macros — proc-macro crate: #[pf_agent] attribute that generates the Spin #[http_component] wiring for WASM agents.
  4. pf-types — lightweight naming and validation primitives (AgentSlug, normalize_name). Intentionally minimal for broad dependency use.
  5. pf_config — layered configuration loader (JSON files → .env → OS environment → optional Cargo.toml section) with typed deserialization.

Feature flags and public exports are defined in each crate’s Cargo.toml and src/lib.rs. Use those as the source of truth; this site summarizes them and does not replace cargo doc.

agent_sdk Runtime-first SDK facade: AgentBuilder, skills, services, LLM engine, A2A/AG-UI adapters.
agent_core Protocol-agnostic domain types — messages, content parts, task phases, conversation context.
pf_macros Proc-macro: #[pf_agent] attribute for Spin/A2aApp wiring.
pf-types Lightweight naming and validation primitives (AgentSlug, normalize_name).
pf_config Layered configuration loader (JSON, env, dotenv, Cargo.toml section).

See also: Guides for step-by-step walkthroughs, and Concepts for architectural background.