promptfleet-agents: Build Rust AI Agents for WASM and Native

Overview

PromptFleet Agents is a collection of independent Rust crates for building AI agents, LLM applications, and agentic workflows. It supports WASM deployments with Fermyon Spin or SpinKube and native applications with Axum and Tokio. The ecosystem includes Google A2A for agent-to-agent communication, MCP for tool and server integration, and AG-UI for streaming output to user interfaces.

Crates can be used separately or composed through the optional pf_agent_sdk. This makes the project relevant to Rust teams that need a particular capability, such as LLM calls or observability, as well as teams building agents that combine protocols and runtime features.

Key Features

• Provider-agnostic LLM client supporting OpenAI, Anthropic, and Azure, with streaming on native and buffered responses on WASM.
• Google A2A v1.0 protocol crates for agent discovery, task handling, JSON-RPC, and streaming.
• MCP support for tool discovery and execution, memory operations, and OAuth 2.1 Bearer authentication.
• AG-UI SSE streaming and an option to serve AG-UI and A2A from one native Axum host.
• LLM tool registry and a #[llm_tool] macro that generates JSON Schema from Rust functions.
• Observability crates for structured logging, OpenTelemetry, and Prometheus.
• Compile-time target adaptation for WASM and native in most library crates.

Use Cases

• Rust developers building an LLM-backed application who want to add provider access, token budgeting, or function calling without adopting a full agent SDK.
• Teams implementing Google A2A agents that need discovery and task endpoints on native or WASM runtimes.
• Agent developers connecting to MCP servers or exposing MCP tools from a Rust application.
• Teams serving both machine-to-machine agent traffic and user-facing streamed output from a native service.
• Operators adding structured logs, traces, or Prometheus metrics to Rust agent workloads.

Getting Started

For a native A2A agent, add the SDK and Tokio to Cargo.toml:

[dependencies]
agent_sdk = { package = "pf_agent_sdk", version = "0.1", features = ["a2a-agent"] }
tokio = { version = "1", features = ["full"] }
anyhow = "1"

See the README and documentation for crate-specific setup and examples.

Alternatives

• rig: Rig focuses on shared interfaces for model providers, vector stores, and agent workflows, while PromptFleet emphasizes A2A, MCP, and AG-UI building blocks across WASM and native targets.
• adk-rust: ADK-Rust provides a broader agent framework with providers, tools, memory, workflows, and voice, while PromptFleet centers on composable protocol crates and WASM deployment.

Considerations

• The repository is marked quiet, with no commits in the last 90 days. It has had releases within the last year, but recent issue and pull request activity is also absent.
• The project has two contributors, with nearly all recorded contributions concentrated in one person. That concentration may affect maintenance continuity.
• The stated Rust requirement is 1.85. There is no Docker, Compose, or Helm requirement in the provided project facts.
• The project facts do not report tests, so teams should verify their needed behavior and target compatibility themselves, especially across WASM and native builds.
• AG-UI is described as native-only, and pf_test_harness is native-only. The README also cautions that some examples and the test harness should be excluded from blanket WASM workspace checks.
• Apache-2.0 is the stated license; review its terms for your distribution and usage needs.