Rust engine that runs agent workflows from a single YAML graph
It exists to give agent workflows that mix LLM calls, database queries, and MCP tools a single declarative execution engine instead of bespoke glue code.

What it does
riceprompt-engine is a Rust library that ingests a YAML workflow definition—nodes, edges, prompts, data sources—and executes the resulting directed graph. It resolves dependencies, dispatches requests to over a dozen LLM providers, runs Rhai scripts, queries PostgreSQL or MongoDB, invokes MCP tools, and orchestrates multi-agent plans through a supervisor node. The ExecutionResult can optionally carry the original YAML topology alongside runtime outputs, so downstream tools can reconstruct the full run from one self-describing file.
The interesting bit
A harness layer injects workflow-wide instructions (think CLAUDE.md) into every generate node, giving persistent memory across the graph without manual plumbing. Even more unusually for this space, the engine supports checkpoint and resume, treating long-running agent workflows as durable jobs rather than stateless requests.
Key highlights
- Entire workflow—graph, prompts, providers, data connectors—lives in one declarative YAML file;
docs/FLOW_SPEC.mdholds the authoritative spec. - First-class streaming, tool calling, and structured output across OpenAI, Anthropic, Gemini, DeepSeek, Qwen, and other providers.
- Node types include Rhai scripting (
transform), iteration (iterator), multi-agent routing (supervisor), subgraphs, and MCP tool nodes. - Built-in connectors for PostgreSQL, MySQL, MongoDB, Redis, Qdrant, S3-compatible storage, and REST APIs.
- Powers the visual IDE RicePrompt, which exports the same YAML the engine consumes.
Caveats
- The API is at
0.1.xand may change between minor versions while the YAML spec stabilizes; the README advises pinning an exact version. - A user-facing usage guide is still listed as forthcoming, so the raw
docs/FLOW_SPEC.mdis currently the primary reference.
Verdict
Worth evaluating if you are building agent infrastructure in Rust and want a checkpoint-aware, declarative alternative to ad-hoc scripting. Give it a pass if you need a frozen v1.0 API today or if your needs are met by a handful of sequential await calls.
Frequently asked
- What is jieyefriic/rp-engine?
- It exists to give agent workflows that mix LLM calls, database queries, and MCP tools a single declarative execution engine instead of bespoke glue code.
- Is rp-engine open source?
- Yes — jieyefriic/rp-engine is open source, released under the Apache-2.0 license.
- What language is rp-engine written in?
- jieyefriic/rp-engine is primarily written in Rust.
- How popular is rp-engine?
- jieyefriic/rp-engine has 1.2k stars on GitHub.
- Where can I find rp-engine?
- jieyefriic/rp-engine is on GitHub at https://github.com/jieyefriic/rp-engine.