Spin Up E2e Stack

Spin up the full E2E stack

Three services, three terminals. Run in this order so each service is ready before the next one connects.

                       ┌────────────────────┐
┌─────────────┐        │  relayer (:3000)   │ ──▶ Sui RPC (gRPC) + Walrus testnet
│  chat-app   │ ─────▶ └────────────────────┘
│  (:5173)    │ ─────▶ Sui GraphQL (group discovery)
└─────────────┘
                       ┌────────────────────┐
                       │ walrus-indexer     │ ──▶ Sui RPC (gRPC) + Walrus testnet
                       │ (:3001, optional)  │     (consumed by SDK RecoveryTransport,
                       └────────────────────┘      not by chat-app)

Prereqs

• Rust stable, Node + pnpm >=10.17.0, Docker (optional, only if running relayer in container).
• Working internet to reach Sui testnet RPC and Walrus testnet endpoints.

Use testnet, not mainnet, for this whole flow. This skill assumes a dev wallet funded from the testnet faucet. Every "Create a group" / "Send a message" in the smoke test below mints real on-chain state on whichever network your relayer + chat-app are pointed at. Mainnet group/message creation costs real SUI, persists permanently, and may be visible to real users if your chat-app's group-discovery surface exposes it. Keep GROUPS_PACKAGE_ID (relayer) + VITE_* package configs (chat-app) on testnet for development.

Want fully-local (localnet with your own contracts)? This skill is testnet-first. On a localnet + testnet-Seal hybrid, message decryption does not work (Seal key servers run against testnet and can't authorize localnet group objects), GraphQL must be locally enabled for group discovery, and Walrus is testnet-only. If you need an all-local stack with working encryption, see spin-up-local-devstack, which runs a local Seal key server via @mysten-incubation/devstack. For a manual localnet bring-up without devstack, note these gotchas surfaced in practice:

• Relayer gRPC on localnet is the fullnode RPC port :9000, not :9124 (the Consistent Store, which 404s on subscribe_checkpoints).
• Publishing the canonical package to localnet needs the suins MVR dep patched to git + sui client test-publish --publish-unpublished-deps (mirror ts-sdks/packages/sui-stack-messaging/test/helpers/localnet/localnet-setup.ts).
• The chat-app must pass packageConfig.permissionedGroups (the localnet sui_groups IDs) or group ops auto-detect testnet and fail. The code reads VITE_MESSAGING_* (not the stale VITE_MESSAGING_GROUPS_* in .env.example).

Recommended path: Docker for the two backend services

Prefer Docker for the relayer and the walrus-indexer. Both ship Dockerfiles; containers avoid host-toolchain friction. Steps 1–3 below are the no-Docker (host) alternative. The chat-app is a Vite dev server with no Dockerfile, so it always runs on the host (Step 3).

# relayer  (terminal 1)
cd relayer && cp .env.example .env          # set GROUPS_PACKAGE_ID (testnet id below)
docker compose up -d                         # :3000

# walrus-indexer  (terminal 2)
cd ../walrus-discovery-indexer && cp .env.example .env   # NETWORK=testnet (default)
docker build -t walrus-indexer .             # on pnpm v11 see the note below if this fails
docker run --rm -p 3001:3001 --env-file .env walrus-indexer   # :3001

# chat-app  (terminal 3, host) — see Step 3

If docker build fails with ERR_PNPM_IGNORED_BUILDS (esbuild) or ERR_PNPM_LOCKFILE_CONFIG_MISMATCH (overrides): you're on pnpm v11, which this repo isn't configured for yet (it targets pnpm >=10.17.0). Simplest fix — change corepack prepare pnpm@latest to corepack prepare pnpm@10 in walrus-discovery-indexer/Dockerfile, then rebuild. Full explanation + the pnpm migration guide in Troubleshooting.

Step 1 — Relayer (terminal 1)

See spin-up-relayer for full details.

cd relayer
cp .env.example .env
# set GROUPS_PACKAGE_ID for testnet:
#   0xba8a26d42bc8b5e5caf4dac2a0f7544128d5dd9b4614af88eec1311ade11de79
cargo run

Wait for: listening on 0.0.0.0:3000 and the membership-sync log line.

Step 2 — Walrus discovery indexer (terminal 2)

Docker (above) is the recommended way to run this service. This repo targets pnpm >=10.17.0; on pnpm v11 the host pnpm install below trips the esbuild build gate (ERR_PNPM_IGNORED_BUILDS). Run pnpm install --ignore-scripts instead — no change to your global toolchain. (Don't corepack … --activate a different pnpm globally just for this; it changes the default for all your projects.) See Troubleshooting.

cd walrus-discovery-indexer
pnpm install
cp .env.example .env
# set NETWORK=testnet (already default)
pnpm dev                # tsx watch on src/index.ts

Listens on :3001 by default (PORT env var). Subscribes to Sui checkpoint stream and inspects certified Walrus blobs. No database — in-memory store.

Health: curl http://localhost:3001/health (returns last processed checkpoint + archived-blob discovery stats — see walrus-discovery-indexer/src/api.ts:62).

Step 3 — Chat-app (terminal 3)

The chat-app can resolve the SDK two ways. Almost everyone wants Mode A.

Mode A — published SDK (default; use this unless you are changing the SDK itself). Depend on the npm package: no ts-sdks/ build, no link:, and the canonical workspace is never touched. In chat-app/package.json set the dependency to the published version (currently 0.0.2):

// chat-app/package.json
"@mysten/sui-stack-messaging": "0.0.2"   // was: link:../ts-sdks/packages/sui-stack-messaging

cd chat-app
cp .env.example .env       # defaults point to localhost relayer + testnet Walrus
pnpm install               # no build:deps needed
# pnpm v11: if this trips the esbuild gate (ERR_PNPM_IGNORED_BUILDS, via vite), add the flag:
#   pnpm install --ignore-scripts          (see Troubleshooting)
pnpm dev                   # vite on :5173

Mode A only swaps the dependency; the build script still chains build:deps ("build": "pnpm build:deps && tsc -b && vite build"), which rebuilds the canonical ts-sdks/ SDK you no longer depend on and fails on pnpm v11 (the ts-sdks/ --frozen-lockfile install hits ERR_PNPM_*). pnpm dev is unaffected — it doesn't run build:deps. For a production build in Mode A, drop build:deps from the build script in your fork ("build": "tsc -b && vite build").

Mode B — linked workspace SDK (repo maintainers, or open-source contributors trying changes to the canonical ts-sdks). Not for app builders — only use this if you are developing @mysten/sui-stack-messaging itself inside this repo. Keep the link:../ts-sdks/... dependency and build it first:

cd chat-app
cp .env.example .env
pnpm install
pnpm build:deps            # builds @mysten/sui-stack-messaging from the ts-sdks/ workspace
pnpm dev                   # vite on :5173

build:deps runs pnpm install --frozen-lockfile inside ts-sdks/, which on pnpm v11 fails with ERR_PNPM_* (the canonical workspace still uses pnpm-v10 config, and --frozen surfaces the overrides mismatch). Don't edit ts-sdks/ (canonical) and don't globally switch your pnpm. Build the SDK with a project-scoped pnpm 10 (e.g. npx pnpm@10 …), or just use Mode A. See Troubleshooting.

Open http://localhost:5173. Connect a wallet (testnet) with some testnet SUI for gas.

Env wiring at a glance

The chat-app does not interact with the walrus-discovery-indexer at all — there is no VITE_INDEXER_URL in chat-app/.env.example and no indexer references in chat-app/src/. The indexer is started in this skill purely so the full stack runs in parallel; you can skip Step 2 entirely if you only want to use the chat-app.

Two distinct concerns that both contain the word "discovery" but are unrelated:

• Group Discovery (finding PermissionedGroup<Messaging> objects the user is a member of) — handled by the chat-app via Sui GraphQL queries against VITE_SUI_GRAPHQL_URL. See docs/sui-stack-messaging/GroupDiscovery.md.
• Archived-message recovery (locating Walrus blobs/quilts the relayer wrote during archival) — what walrus-discovery-indexer is for. The SDK's RecoveryTransport is the consumer (not the chat-app). See docs/sui-stack-messaging/ArchiveRecovery.md.

Containers — detail and caveats

This expands the recommended Docker path at the top. Both backend services ship Dockerfiles, so you can run the stack without a Rust/Node toolchain on the host — also how you'd mimic a deployment topology or run from CI.

# Relayer — already has docker-compose
cd relayer
cp .env.example .env                        # set GROUPS_PACKAGE_ID
docker compose up -d                        # :3000, healthcheck on /health_check (see docker-compose.yml)

# Walrus indexer — single Dockerfile (no compose)
cd ../walrus-discovery-indexer
cp .env.example .env                        # set NETWORK=testnet
docker build -t walrus-indexer .            # pnpm v11 fails here; build on pnpm 10 (corepack prepare pnpm@latest -> pnpm@10 in the Dockerfile). See Troubleshooting
docker run --rm -p 3001:3001 --env-file .env walrus-indexer
# image exposes :3001 with HEALTHCHECK on /health (see walrus-discovery-indexer/Dockerfile)

The chat-app stays as pnpm dev for hot reload during development. If you want it containerized too, build once with pnpm build and serve dist/ behind any static webserver — there's no Dockerfile shipped for it.

Caveats:

• Inter-container networking is not pre-wired. If you put both services on the same Docker network, the chat-app's VITE_RELAYER_URL (set at build time) and the relayer-to-Sui gRPC connection both have to be configured for the topology you choose. Running the chat-app on the host while the backend services are containerized is the simplest combination.

Build the SDK once when working across packages

cd chat-app
pnpm build:deps          # rebuilds the linked SDK package
# vite will hot-reload after this

chat-app/package.json links the SDK as link:../ts-sdks/packages/sui-stack-messaging

Quick smoke test

1. Connect wallet in chat-app (top right).
2. Create a group → wait for tx confirmation.
3. Send a message → relayer logs POST /messages 200.
4. Refresh, message reappears (read path also OK).

Troubleshooting

• chat-app shows "relayer unreachable" — relayer not running, or CORS misconfig (relayer permits localhost by default; check tower-http cors layer in relayer/src/main.rs).
• After relayer restart, the chat-app still lists old groups but interactions fail — expected with the reference relayer. Group existence is on-chain, so the chat-app's GraphQL-backed discovery still surfaces every group the wallet is a member of. The relayer, however, does not backfill: its membership store starts empty on each launch and is rebuilt only from Sui gRPC checkpoints emitted after startup. Until membership events for a given group fire again on-chain, the relayer will reject reads/writes against it. Workaround: implement checkpoint backfill in your fork — see develop-relayer and the same caveat in develop-on-sui-stack-messaging.
• pnpm v11 fails a fresh install — ERR_PNPM_IGNORED_BUILDS (esbuild) and/or ERR_PNPM_LOCKFILE_CONFIG_MISMATCH (overrides), on the indexer and chat-app (relayer is Rust, unaffected). pnpm v11 no longer reads this repo's package.json pnpm config. Quick fix: Docker → pin pnpm@10 in the Dockerfile; host → pnpm install --ignore-scripts (don't globally switch pnpm). Root cause, both errors, all remedies, and the pnpm migration guide: docs/pnpm-v11-troubleshooting.md.
• Wallet has no SUI — fund via https://faucet.testnet.sui.io

Want to deploy each piece separately?

• Relayer alone: spin-up-relayer.
• Customize relayer behavior: develop-relayer.
• Customize walrus indexer: develop-walrus-indexer.
• Add Move-side functionality: extend-smart-contracts.