PolicyEngine Chat

A chat interface for the PolicyEngine agent. Ask questions about UK or US tax and benefit policy, calculate household impacts, and analyse economy-wide reforms.

Setup

Prerequisites

• Bun runtime
• Modal account (for the agent)
• Supabase project (already deployed at xieouadfboiipmwqrhyg)

Local development

1. Install dependencies:

bun install

2. Create .env.local (or use the existing one):

cp .env.local.example .env.local

3. Start the dev server:

bun dev

Open http://localhost:3000 to use the chat.

Deploying the Modal agent

The agent runs on Modal and stores logs in Supabase. To redeploy:

cd modal_agent
modal deploy agent.py

The Modal function needs two secrets:

• anthropic-api-key - your Anthropic API key
• policyengine-chat-supabase - Supabase URL and service key

Environment variables

Variable	Description
NEXT_PUBLIC_SUPABASE_URL	Supabase API URL
NEXT_PUBLIC_SUPABASE_ANON_KEY	Supabase anonymous key
SUPABASE_SERVICE_KEY	Supabase service role key (server-side only)
NEXT_PUBLIC_API_BASE_URL	PolicyEngine API URL (default: https://v2.api.policyengine.org)

Database schema

Three tables:

• threads: Chat sessions with title and timestamps
• messages: User and assistant messages
• agent_logs: Streaming logs during agent execution

See supabase/migrations/001_initial.sql for the full schema.

Architecture

┌─────────────┐     ┌──────────────┐     ┌─────────────┐
│   Next.js   │────▶│    Modal     │────▶│  Supabase   │
│   Frontend  │     │    Agent     │     │   Database  │
└─────────────┘     └──────────────┘     └─────────────┘
       │                   │                    │
       │                   ▼                    │
       │           ┌──────────────┐             │
       │           │ PolicyEngine │             │
       │           │    API v2    │             │
       │           └──────────────┘             │
       │                                        │
       └────────────────────────────────────────┘
                    (realtime updates)

1. User sends message → saved to Supabase
2. Next.js API spawns Modal agent
3. Agent calls PolicyEngine API, streams logs to Supabase
4. Agent saves final response to Supabase
5. Frontend receives updates via Supabase realtime