v0.1.2 — Mainnet deployment + all four npm packages

MemForks v0.1.2

Git for AI agent memory. Branch-aware, verifiable, mergeable memory for autonomous agents — built on Sui + MemWal.

---

What's new since v0.1.1

• Mainnet deployment — contracts are live on Sui mainnet
• All four @memfork/* packages published to npm
• Framework adapters fixed for ai@4.x and @langchain/langgraph-checkpoint@1.0.x API changes

---

Contract deployments

Network	Package ID	Registry ID
mainnet	0xcee7a6fd8de52ce645c38332bde23d4a30fd9426bc4681409733dd50958a24c6	0x0da982cefa26864ae834a8a0504b904233d49e20fcc17c373c8bed99c75a7edd
testnet	0xcf6ad755a1cdff7217865c796778fabe5aa399cb0cf2eba986f4b582047229c6	—

---

npm packages

Package	Version	Install
@memfork/core	0.1.1	npm i @memfork/core
@memfork/cli	0.1.1	npm i -g @memfork/cli
@memfork/vercel-ai	0.1.0	npm i @memfork/vercel-ai
@memfork/langgraph	0.1.0	npm i @memfork/langgraph

---

Quick start

npm install -g @memfork/cli

memfork init --quick       # keygen → faucet → MemWal account → tree (~30s)
memfork doctor             # verify everything is working
memfork install cursor     # wire into Cursor

Vercel AI SDK

import { withMemForks } from "@memfork/vercel-ai";
import { openai } from "@ai-sdk/openai";

const model = withMemForks(openai("gpt-4o"), { branch: "feature/my-feature" });

LangGraph

import { createMemForksCheckpointer } from "@memfork/langgraph";

const checkpointer = await createMemForksCheckpointer();
const app = new StateGraph(MessagesAnnotation)
  .addNode("agent", myNode)
  .compile({ checkpointer });

---

Known gaps (coming in v0.2.0)

• memfork proposals — real table view (currently redirects to memfork ui)
• memfork install-hooks — git post-checkout auto branch-sync
• mem.on('commit', ...) event subscriptions on the SDK client
• memfork migrate — import from LangGraph / JSON into an existing tree
• CrewAI adapter (memfork-crewai)
• Conformance test suite

v0.1.1 — Developer surface + team onboarding

MemForks v0.1.1 — Pre-release

Git for AI agent memory. Branch-aware, verifiable, mergeable memory for autonomous agents — built on Sui + Walrus + MemWal.

---

What works today

Core protocol (Phases 0–2 — complete)

• Move package live on Sui testnet: memforks::tree, memforks::acl, memforks::resolver
• Full branch → commit → recall → propose_merge → attest → finalize_merge ceremony verified on testnet
• Model A architecture: commits are off-chain Walrus blobs (fast, free); merges are Sui transactions (settled, auditable)
• 5 resolver kinds: LastWriteWins, Union, JuryReconcile, LlmReconcile, Sequence/And combinators
• Ed25519 attestation verification on-chain
• Package ID: 0x080722f5b7025679aa17792a3b07ef9b875b4ad3cee7640ecf9b8b7abd5b5347
• Reference tree: 0xeb88a31b9ef8c015e0182929c6b499126e176939ccfe5fd419dd8e1b35bea93c

SDK — @memfork/core

• MemForksClient / MemoryClient (alias) with branch, commit, recall, proposeMerge, finalizeMerge, grantDelegate, revokeDelegate, createResolver, waitForFinalization
• Zero-arg connect(): auto-reads .memfork/config.json + ~/.memfork/credentials.json + MEMFORK_* env vars
• resolvers.{lastWriteWins, union, jury, llmReconcile, and, sequence} BCS config builders
• perms / PERM bitmask constants
• MemForksIndexer — event-driven branch + merge state tracking

CLI — memfork

Command	Description
memfork init [--quick]	Automated provisioning: keygen → faucet → MemWal account → tree (~30s)
memfork join	Team member onboarding — generates keys, prints grant commands for owner
memfork doctor	Health check — config, credentials, Sui RPC, MemWal, delegate status
memfork install cursor|codex	Wires MemWal MCP + agent rules into Cursor or Codex
memfork branch <name>	Create a branch (on-chain Sui tx)
memfork checkout <name>	Switch active branch
memfork commit -m <msg> --facts ...	Manually anchor facts as a Walrus blob
memfork diff <a> <b>	Semantic fact diff between two branches
memfork merge <from> <into> --resolver <id>	Propose a merge
memfork grant / grant-memwal / revoke	Access control
memfork ui [--share]	DAG visualizer (local or Walrus Site)

Framework adapters

• @memfork/langgraph — BaseCheckpointSaver; each thread maps to a branch
• @memfork/vercel-ai — withMemForks middleware; recall before generate, commit after

Coding-agent plugins

• Cursor — MemWal MCP (Streamable HTTP) + memforks.mdc rule. memfork install cursor auto-configures ~/.cursor/mcp.json
• Codex — same pattern. memfork install codex auto-configures ~/.codex/config.toml

---

Quick start

npm install -g @memfork/cli

memfork init --quick       # keygen → faucet → MemWal account → tree (~30s)
memfork doctor             # verify everything is working
memfork install cursor     # wire into Cursor

Team member onboarding:

# Teammate (after cloning repo):
memfork join

# Owner runs:
memfork grant --agent <address>
memfork grant-memwal --agent <address> --pubkey <hex>

# Teammate confirms:
memfork doctor

---

Known gaps (coming in v0.2.0)

• memfork proposals — currently redirects to memfork ui; real table view coming
• memfork install-hooks — git post-checkout auto branch-sync not yet built
• createMemForksCheckpointer() no-arg form for LangGraph
• mem.on('commit', ...) event subscriptions on the SDK client
• memfork migrate — import from LangGraph / JSON into an existing tree
• CrewAI adapter (memfork-crewai)
• Conformance test suite

---

Docs

• Developer Guide
• Git → MemForks comparison
• Architecture (Model A)

---

@memfork/vercel-ai v0.1.0

@memfork/vercel-ai v0.1.0

Drop-in Vercel AI SDK middleware that gives any model branch-aware, on-chain memory via MemForks.

Install
```bash
npm install @memfork/vercel-ai @memfork/core ai
```

Usage
```typescript
import { withMemForks } from "@memfork/vercel-ai";
import { openai } from "@ai-sdk/openai";

const model = withMemForks(openai("gpt-4o"), { branch: "feature/my-feature" });
```

What it does

• Before generating — recalls top-N semantically relevant facts from the branch and injects them into the system prompt
• After generating — fire-and-forget commit anchors the response on-chain with full provenance
• Works with generateText, streamText, generateObject
• branchFromContext option for per-request dynamic branch resolution
• Zero-config: reads credentials from ~/.memfork/credentials.json or MEMFORK_* env vars

Links

• npm
• README

@memfork/langgraph v0.1.0

@memfork/langgraph v0.1.0

Drop-in LangGraph checkpointer that stores every graph state snapshot on Sui (via MemWal). Each LangGraph thread maps to a MemForks branch — enabling cross-agent memory merges.

Install
```bash
npm install @memfork/langgraph @memfork/core
```

Usage
```typescript
import { createMemForksCheckpointer } from "@memfork/langgraph";

const checkpointer = await createMemForksCheckpointer({
treeId: process.env.MEMFORK_TREE_ID!,
signer: process.env.MEMFORK_PRIVATE_KEY!,
memwal: {
accountId: process.env.MEMFORK_MEMWAL_ACCOUNT!,
delegateKey: process.env.MEMFORK_MEMWAL_KEY!,
},
});

const app = new StateGraph(MessagesAnnotation)
.addNode("agent", myNode)
.compile({ checkpointer });
```

What it does

• Implements BaseCheckpointSaver — drop-in replacement for MemorySaver
• Each put() commits serialised graph state to MemWal and anchors the blob on-chain
• getTuple() / list() restores state via semantic recall
• Thread IDs map to branches: thread/<thread_id>
• proposeMerge() for cross-agent state reconciliation via on-chain resolvers
• Zero-config: reads credentials from ~/.memfork/credentials.json or MEMFORK_* env vars

Peer dependencies

• @langchain/core >=0.3.0
• @langchain/langgraph >=1.0.0
• @langchain/langgraph-checkpoint >=1.0.0

Links

• npm
• README

v0.1.0 — Core protocol

MemForks v0.1.0 — Core protocol

Initial release. Full branch → commit → propose_merge → attest → finalize_merge ceremony working on Sui testnet.

• Move package deployed: memforks::tree, memforks::acl, memforks::resolver
• @memfork/core SDK
• @memfork/langgraph + @memfork/vercel-ai adapters
• Cursor + Codex plugins
• memfork CLI: init, doctor, install, status, log, recall, commit, merge, ui