flowcraft
Production-grade Go SDK for building AI agents with long-term memory, knowledge retrieval, and voice — runnable as a library, a daemon, or a real-time pipeline.
What is flowcraft?
Production-grade Go SDK for building AI agents with long-term memory, knowledge retrieval, and voice — runnable as a library, a daemon, or a real-time pipeline.
Overview of flowcraft
What it is
FlowCraft is a production-grade Go SDK for building AI agents with long-term memory, knowledge retrieval, and voice. It can be run as a library, a daemon, or a real-time pipeline. The project is a modular Go workspace that avoids tying application code to one model provider or execution model. It includes multiple modules: core (agent execution, graph, tool, model, memory contracts, event bus, telemetry, etc.), craft (application assembly, plugin host, lifecycle management), driver (provider adapters for OpenAI, Anthropic, ByteDance, MiniMax, DashScope, TypeSafe), backends/* (platform-specific implementations like SQLite checkpoints and long-term memory), and examples (forge for a local workspace demo, anvil for a minimal craft host application).
Key features
- Modular architecture with independently versioned Go modules; applications adopt only the layers they need.
- Long-term memory architecture with contracts (
ContextProvider,TurnSink,DocumentSink, etc.) and pluggable implementations. - Multi-provider support via driver adapters (OpenAI, Anthropic, ByteDance, MiniMax, DashScope, TypeSafe); any endpoint speaking the same wire family can be used.
- Plugin system via MCP (Model Context Protocol): plugins run as MCP servers, host primitives are exposed as MCP tools for plugins.
- Multiple execution modes: embeddable as a library, run as a daemon, or as a real-time pipeline.
- Built-in workspace demo (
examples/forge) with interactive TUI, scripted tests, and raid × persona simulations.
Use cases
- Building AI agents with long-term memory and knowledge retrieval.
- Creating local interactive workspaces for AI application development and testing.
- Assembling production-grade AI services that can switch between model providers without code changes.
- Developing real-time voice-enabled AI pipelines.
- Embedding AI agent capabilities into existing Go services.
Getting started
bash # Run the forge workspace demo cd examples/forge go run . help # Create a workspace from the werewolf scenario and run a scripted test go run . workspace create --config werewolf --workspace ./workspace go run . test -test werewolf/opening_setup # Run the craft example (no provider or network needed) cd examples/anvil go run . For embedding FlowCraft in a Go service, see the README for code examples using core/deploy, core/runtime, and core/runtime/session.
License and pricing
The project is licensed under the MIT License and is open source (no pricing stated).
Based on the project's public README and repository metadata.