Anthony Sun
MY APPS

Apps I've Built

Each one is a real product, not a demo.

How it's built↓
Landfox home page — address search beside a live product preview of planning, hazard and school data

Landfox

LIVE

Know the land before you commit.

Property intelligence for any Victorian or NSW address — planning zones, overlays, flood and bushfire risk, schools, crime and market data from official state sources, with an AI planner you can ask about any of it. Free to start; PDF reports and an MCP server for AI agents on paid plans.

Open App
HOW IT'S BUILT

One stack, four layers

Every app sits on the same agent framework, services and data layer. Each layer only talks to the one below it.

  1. 01 · Application

    AI-native apps

    Web apps where every user action is also an API call, so an agent can do anything a person can. Streaming answers, live plans and approval steps are part of the interface.

    • Next.js
    • React
    • TypeScript
    • Tailwind CSS
    • Clerk
    • MapLibre
    • Stripe
  2. 02 · Agents

    as_agents harness

    My own agent framework on LangGraph. Each agent has an identity, skills, tools and memory. It can plan, hand work to other agents and stop for human approval.

    • Persistent memory, with every thread checkpointed in Postgres
    • Agentic flows: multi-step plans, delegation and background jobs
    • Context optimisation: skills load only when needed and long threads are compressed, so each call costs fewer tokens and the model sees less noise to hallucinate from
    • as_unifiedllm: one model layer over OpenAI, DeepSeek, OpenRouter and local models, with token accounting
    • Python
    • LangGraph
    • LangChain
    • as_unifiedllm
    • FastAPI
    • MCP
    • Agent Skills
    • Ollama
  3. 03 · Services

    Tool services

    Standalone FastAPI services the agents call as tools. Apps can call them directly too.

    • as_extractor: PDFs, Office files, images, audio and video into structured JSON
    • as_property: Victorian property and planning data, with PDF reports
    • Web search and scrape: pages turned into compact, sourced facts
    • FastAPI
    • PyMuPDF
    • faster-whisper
    • Firecrawl
    • Camoufox
    • WeasyPrint
  4. 04 · Data

    Data and scheduling

    Where state lives, and what keeps work running when nobody is watching.

    • as_vectordb: semantic search over everything ingested
    • Postgres on Supabase for app data, agent history and token telemetry
    • Scheduler and queues that run agent tasks on their own
    • Qdrant
    • sentence-transformers
    • PostgreSQL
    • Supabase
    • as_queues

Built for agents, not just people

The framework and the apps were designed with agents in mind from the start. Services speak MCP and an OpenAI-compatible API, so they plug into the agent harness you already use.

Works with

More integrations coming soon.

  • Claude Code
  • Codex CLI
  • OpenCode
  • Hermes Agent
  • LangGraph
  • CrewAI
  • Windsurf
  • Cursor
  • GitHub Copilot
  • Gemini CLI