Phoenix Tree AI
Menu
Get updates

Phoenix Tree AI

Open systems for work that has to add up.

Evidence-led systems for independent firms. Built to keep sources, calculations and gaps attached to the work.

In development

Phoenix Tree AI is a Hong Kong-based AI company founded and operated by Simon Lin.

2 products, built around the work that small firms already do.

Global market research and multi-channel commerce use different data. Both become unreliable when a number loses its source, a gap gets hidden or an automated step cannot be checked later.

Current products

What the current builds can do

All 2 products run locally today. None has a supported public release. The details below come from the current repositories and working interfaces.

01
Global investment research In development R8 continuity complete

Global Markets Research

Working title. The public product name is not locked.

A local research workspace for market evidence, monitored claims and research cases that keep changing after the first report is written.

See what exists today

Verified now

  • 9 provider-neutral research workflows passed the R7 release acceptance run on 31 August 2026
  • The current workspace covers US, Hong Kong, Korea, Taiwan and Japan alongside digital-asset and derivatives research paths
  • R8 preserves research dossiers, evidence changes and frozen earlier decisions instead of rewriting history
02
Multi-channel commerce In development P24 technical release candidate

Open Commerce Workbench

Working title. The public product name is not locked.

A local operating desk for inventory, orders, profit, exceptions and source evidence across more than one sales channel.

See what exists today

Verified now

  • The P24 local workspace has eight primary operating views and a context-aware Operations Inspector
  • Amazon Orders and Finances passed a real read-only account workflow
  • Inventory, order and profit calculations remain deterministic and keep their evidence links

How we build

The model is not the source of truth

01

Local before cloud

The working record stays on the operator's machine. A useful core remains even when no model is connected.

02

Deterministic before generative

Ledgers, calculations, filters and validation run as code. The model investigates and explains; it does not get to invent the numbers.

03

Evidence before action

A result keeps the source record, the transformation and the gap together. Writes remain behind an explicit approval step.

Read the full build method

Product truth

Target coverage is not shipped coverage

The finance product has passed 9 release-acceptance workflows, but its market coverage is not equally deep everywhere. The commerce product has a real Amazon read path and a technical release candidate, but the other marketplace adapters have not passed live-account verification. Remote writes remain disabled.

Building toward

Global Markets Research

  • Global equities, digital assets and derivatives through market-specific evidence paths
  • Living research cases that track source changes, stale claims and review decisions
  • A local workspace for holdings, watchlists, reports and source health

Not available yet

  • The public product name is not locked
  • Coverage is not equally deep across every market and asset class
  • The portfolio screens are not a brokerage connection
  • There is no supported public release yet

Building toward

Open Commerce Workbench

  • One operating view for exceptions, inventory, orders, profit, channels and evidence
  • Read-only marketplace connections with deterministic local ledgers
  • Optional agent investigation with human approval before any external write

Not available yet

  • The public product name is not locked
  • Shopify, eBay, Etsy and Walmart are contract implementations, not live account verification
  • Remote marketplace writes remain disabled
  • Independent seller adoption and a signed, notarized desktop release are not verified

Before launch

Follow the build, not a promise date.

Get product updates