MemForks Documentation

Git for AI agent memory

Branch-aware, verifiable, mergeable memory anchored on Sui. Fork an experiment, let it diverge, diff what it learned, then merge trusted knowledge back into main.

Get started →Learn the model

Verifiable

Every memory is anchored on Sui and stored in Walrus — provenance you can audit, not trust blindly.

Branch-aware

Agents inherit memory along lineage. Branches diverge cleanly and stay queryable, never contaminated.

Mergeable

Promote what an experiment learned back to main through governed, reviewable merges — not blind writes.

Explore the docs

01

Get Started

Install the CLI, provision a memory tree, verify setup, and make your first branch commit.

02

Concepts

Understand MemWal vs MemForks, branch lineage, commits, diffs, and merge governance.

03

Architecture

Explore the Sui, Walrus, MemWal, SDK, CLI, and adapter layers that power the system.

04

Core SDK

Use @memfork/core directly for branch, commit, recall, delegate, and merge flows.

05

Vercel AI SDK

Add branch-aware recall and auto-commit to chat, streaming, and structured generation.

06

LangGraph

Persist graph checkpoints to MemForks and map each LangGraph thread to a memory branch.

07

CLI

Provision trees, inspect status, create branches, recall facts, commit memory, and propose merges.

08

Examples

Run the Next.js chat app and LangGraph research workflow to see MemForks in practice.

09

Operations

Configure hosted apps, troubleshoot credentials, and run gas sponsorship infrastructure.

Memory with lineage

MemWal stores encrypted memories and recalls them by meaning. MemForks adds the version-control layer: which branch learned a fact, what it inherited, and how knowledge should merge back.

main:          c1 ── c2 ── c3 ─────────── c7
                          \                /
hypothesis-a:              c4 ── c5 ──────
                               \
hypothesis-b:                   c6          abandoned, still queryable