LangGraphGo

LangGraphGo is a Go implementation of the core ideas behind LangGraph: a graph-based runtime for building deterministic, stateful LLM workflows with checkpointing, interrupts/resume, and streaming.

This repository focuses on:

• A Pregel-style execution engine (step-based scheduling)
• Durable checkpointing (save/restore) and resume semantics
• Tool calling support through provider adapters
• A practical Go-first API surface (while tracking LangGraphJS semantics)

What it does

LangGraphGo lets you define a graph of nodes (functions) connected by edges. Nodes can:

• Read from state channels
• Write updates to channels (with reducer semantics depending on channel type)
• Emit interrupts and resume later using checkpoints
• Stream events/updates during execution

The runtime executes the graph step-by-step, applying writes deterministically and persisting progress via a configurable checkpointer.

Why Go

Porting LangGraph concepts to Go enables benefits that fit production systems:

• Concurrency & parallelism: goroutines make it natural to run independent node work concurrently.
• Lower overhead, simpler deployments: a single static binary is easy to ship.
• Predictable performance: explicit control over memory and concurrency limits.
• Strong typing where it matters: core runtime contracts are explicit, while state payloads remain flexible (any) for ergonomics.

Structural differences vs LangGraphJS (high-level)

This project aims for semantic parity with LangGraphJS, but some surface-area and implementation details differ due to Go idioms:

• Go-first public API: fewer builder/annotation conveniences; focus on explicit types and simple composition.
• Concurrency model: Go may execute work in parallel while preserving deterministic observable semantics.
• Checkpointing contract: checkpointers are pluggable (memory/file/sqlite/postgres/redis/mongo), and checkpoint tuples include pending writes and optional parent linkage.
• Streaming: exposes stream chunks/events using Go channels.

Status

This is an active work-in-progress.

Parity with LangGraphJS is not complete yet. I continue to implement missing pieces when I have time and will keep tightening semantics and public API coverage.

Contributing

Contributions are welcome.

• Open an issue for feature requests / parity gaps.
• PRs for bug fixes, tests, provider adapters, and documentation improvements are appreciated.

License

See the repository license file.