AlphaForge

Live demo: https://alphaforge-five.vercel.app

AlphaForge is an AI trading agent built for hackathon use cases, combining market signals, LLM decision-making, risk controls, paper trading execution, and ERC-8004 on-chain identity.

What it does

Every cycle, the agent:

• fetches market signals from PRISM,
• sends the current market + portfolio context to an LLM,
• applies strict risk checks,
• executes a paper trade through Kraken CLI,
• stores everything in PostgreSQL via Prisma,
• optionally posts attestations and feedback on Sepolia through ERC-8004 flows.

Architecture

Next.js Dashboard
  ├─ UI pages
  ├─ API routes
  └─ agent controls

Agent Loop
  ├─ PRISM market signals
  ├─ LLM decision engine
  ├─ risk engine
  ├─ Kraken paper execution
  └─ PostgreSQL persistence

On-chain Layer
  └─ ERC-8004 identity / validation / reputation on Sepolia

Main flows

1. Market intelligence

The app pulls live market context from PRISM, including price, momentum, and risk-related signals.

2. AI decision engine

The agent sends structured input to the LLM and expects a constrained output like:

• BUY
• SELL
• HOLD

Each decision also includes confidence and reasoning.

3. Risk enforcement

Before any execution, the agent applies hard guardrails:

• stop-loss,
• take-profit,
• max exposure,
• duplicate position prevention.

4. Trade execution

If the action passes the risk engine, AlphaForge executes a paper trade through Kraken CLI.

5. Persistence

Trades, runs, and config are saved with Prisma + PostgreSQL.

6. On-chain reputation

The agent can publish identity and trading-related attestations to Sepolia using ERC-8004-compatible flows.

App routes

• / → live dashboard
• /trades → trade history
• /signals → market signals
• /agent → agent configuration

API routes

• POST /api/agent/start
• POST /api/agent/stop
• GET /api/agent/status
• POST /api/kraken/execute
• GET /api/market/signals

Project structure

app/
  agent/
  api/
  signals/
  trades/

components/
  agent/
  charts/
  dashboard/
  navigation/
  ui/

lib/
  agent/
  kraken/
  onchain/
  prism/
  prisma.ts
  env.ts
  runtime-store.ts

prisma/
  schema.prisma
  migrations/

scripts/
  register-agent.ts
  check-balance.ts
  check-trades.ts
  worker.ts

Tech stack

• Next.js
• React
• Tailwind CSS
• shadcn/ui
• Prisma
• PostgreSQL (pg + @prisma/adapter-pg)
• OpenAI
• PRISM API
• Kraken CLI
• viem
• Sepolia

Deploy on Vercel

1. vercel link --yes --scope <team-slug> --project alphaforge (or create the project in the dashboard).
2. Push environment variables: pnpm run vercel:env
   This reads .env, forces KRAKEN_CLI_SIMULATE=true on Production (Kraken CLI binary is not available in serverless), and ensures CRON_SECRET exists for /api/cron/agent-tick.
3. vercel deploy --prod --yes

Cron on the Hobby plan is limited to once per day (see vercel.json). Set DATABASE_URL (or Prisma Postgres STORAGE_* vars) on Vercel so the app uses one shared Postgres instance.

Database (PostgreSQL)

The app targets PostgreSQL via DATABASE_URL (connection URL also read from STORAGE_PRISMA_DATABASE_URL / STORAGE_POSTGRES_URL when using Vercel’s Prisma Postgres integration).

1. Provision a database (e.g. Prisma Postgres, Neon, or docker compose up -d with docker-compose.yml).
2. Set DATABASE_URL in .env to a postgresql:// or postgres:// URL.
3. Apply migrations: pnpm exec prisma migrate deploy
4. Builds on Vercel run prisma migrate deploy then next build via scripts/vercel-build.cjs. Demo seed is skipped on Vercel by default (so deploys don’t wipe Trade / AgentRun). To seed once, set SEED_DEMO_ON_BUILD=true in the Vercel environment for that deploy, or run pnpm exec tsx scripts/seed-demo.ts locally against your DATABASE_URL.

Optional: Worker + real Kraken CLI (leaderboard / paper)

Vercel runs short-lived functions; use a separate process for the infinite loop and for the kraken binary:

• Locally: pnpm worker with KRAKEN_CLI_SIMULATE=false and KRAKEN_CLI_PATH pointing at your install.
• Container: build Dockerfile.worker, install or mount Kraken CLI in the image, pass the same .env as the Next app.

Hackathon eligibility (e.g. early.surge.xyz) is outside this repo—register your team there per lablab rules.

Local setup

1. Install dependencies

pnpm install

2. Create environment file

Create .env (or .env.local) and add your local values. DATABASE_URL must be a PostgreSQL URL (postgres:// or postgresql://). Use Prisma Postgres, Neon, or run docker compose up -d and point DATABASE_URL at the container.

Example:

DATABASE_URL="postgresql://USER:PASSWORD@HOST:5432/DATABASE?sslmode=require"
TRADING_MODE="paper"

KRAKEN_API_KEY=""
KRAKEN_API_SECRET=""
KRAKEN_CLI_PATH=""
KRAKEN_CLI_SIMULATE="true"

PRISM_API_KEY=""

OPENAI_API_KEY=""
LLM_PROVIDER="openai"
LLM_MODEL="gpt-4o-mini"

AGENT_PRIVATE_KEY=""
AGENT_WALLET_ADDRESS=""
BASE_SEPOLIA_RPC=""
SEPOLIA_RPC_URL=""

3. Run Prisma

pnpm exec prisma migrate deploy

For local schema iteration (creates migration files):

pnpm db:migrate

4. Start the app

pnpm dev

Open http://localhost:4000.

Demo flow

1. Start the app.
2. Open /agent.
3. Start the worker.
4. Watch signals update.
5. Let the loop generate a decision.
6. Inspect trades in /trades.
7. Show persisted data and on-chain registration.

ERC-8004 identity

Field	Value
Agent ID	45
Network	Ethereum Sepolia (chainId 11155111)
Wallet	0x34A0aC099E6812701e970024fff89aDDb95c4426
Registration TX	0x97a063...
Agent card	/agent-registration.json

Contracts (Sepolia):

Contract	Address
AgentRegistry	0x97b07dDc405B0c28B17559aFFE63BdB3632d0ca3
HackathonVault	0x0E7CD8ef9743FEcf94f9103033a044caBD45fC90
RiskRouter	0xd6A6952545FF6E6E6681c2d15C59f9EB8F40FdBC
ReputationRegistry	0x423a9904e39537a9997fbaF0f220d79D7d545763
ValidationRegistry	0x92bF63E5C7Ac6980f237a7164Ab413BE226187F1

Hackathon highlights

• AI-driven trading decisions
• strict risk engine before execution
• paper trading flow through Kraken CLI
• PostgreSQL persistence for reproducible demo
• ERC-8004 identity / validation / reputation integration
• clean dashboard for live inspection

Safety notes

• This project is configured for paper trading during demos.
• Never commit real secrets.
• Never reuse exposed private keys.
• Use fresh credentials before any public demo or deployment.

Submission checklist

• .env.local is not committed
• .env.example is included
• dev.db is ignored
• agent registration JSON is available in public/
• README includes demo steps
• screenshots or GIF added before final submission

License

MIT