AI that helps with real life, on your terms.

GECHEN AI builds products that let AI agents use your knowledge and tools, with sources you can check and access you can revoke. They work with any agent through MCP and open APIs.

Open TripPocket (opens in a new tab) ACP Gateway on GitHub (opens in a new tab)

agent <-> TripPocket illustrative session, not real output

# 1 A person asks, in plain language

you> Can we take the stroller into teamLab Planets?

# 2 The agent calls TripPocket over MCP, with a scoped token

agent-> search_knowledge("stroller teamLab Planets")

Agent Access Token •••••••• scoped, revocable

# 3 TripPocket returns knowledge items, each with its source

trippocket<- 1 knowledge item, from Pocket “Tokyo with kids”

“teamLab Planets is barefoot and knee-deep in water — bring shorts for the kids and leave the stroller at the entrance desk.”

source: teamLab Planets TOKYO — Visitor guide

# 4 The agent answers, and cites where it came from

agent> Leave the stroller at the entrance desk, and bring shorts for the kids: the water is knee-deep. [1]

[1] teamLab Planets TOKYO — Visitor guide

you>

3 products: One in beta, one open source, one in development.
1,300+ commits: Across 4 product repos in 2026.
Taipei, Taiwan: Independent, founder-led and bootstrapped.

What we’re building: software agents can use.

TripPocket and ACP Gateway keep the durable parts — knowledge, permissions, provenance — in the product, and let people bring the agent they prefer. ParkNearby is our map-first iOS app.

TripPocketBetaAn agent-ready travel knowledge layer
ACP GatewayOpen source, MITCoding agents on your machine, used from anywhere
ParkNearbyIn developmentMap-first parking for Taipei and Taoyuan

TripPocket

BetaFlagship

An agent-ready travel knowledge layer.

Travelers save tips, places and sources once. Any authorized AI agent — Claude, or a custom agent — can then read, search and add to that knowledge, with provenance on every item.

Open TripPocket (opens in a new tab)

Two interfaces, one domain model

A remote MCP server and a REST/OpenAPI surface sit over the same model, so agents and apps see the same knowledge.
Scoped Agent Access Tokens

Agents connect with scoped Agent Access Tokens that users can see, and revoke, at any time.
Provenance on every item

Pockets, Places, Sources and Knowledge Items, with source attribution carried on every item.

TripPocket Knowledge view: saved travel tips filtered by tag and Pocket, with the selected tip showing its tags, place, Pockets and the source it came from. Opens the full-size screenshot. — trip-pocket.gechen.orgscreenshot

ACP Gateway

Open source, MITWeb + iOS

ACP Gateway iOS app showing an agent conversation on a phone, with a message composer and model selector. Opens the full-size screenshot.

Run coding agents on your own machine. Use them from anywhere.

Coding agents such as Claude Code run on your machine, and you reach them through an authenticated web console and a native iOS app. Code, credentials and agent state stay on the host.

ACP Gateway on GitHub (opens in a new tab)

See the work, not just the chat

Live diffs, plans and touched files sit beside the conversation.
Approve from web or phone

Permission requests reach you as notifications that deep-link back to the thread.
Branch or hand off

Branch a conversation, or hand it off from one agent to another.
A native iOS console

Built in SwiftUI, with credentials in Keychain and an optional Face ID lock.

ParkNearby

In developmentiOS

Map-first parking for drivers in Taipei and Taoyuan.

Nearby parking, current rates and real-time vacancy from official open data. The age of every data point is shown, so drivers know how fresh it is.

Native SwiftUI
Small API service
Official open data

        |            |           |      
        | [P] 20s    | [P] 1m    |      
        |            |           |      
--------+------------+-----------+------
        |            |           |      
[P] 6m  |            |           |      
        |    (@) you |           |      
        |            | [P] 45s   |      
        |            |           |      
--------+------------+-----------+------
        |            |           |      
        | [P] 3m     | [P] 11m   |      
        |            |           |

Abstract illustration, not product UI. Beside each [P] is the age of its data.

Claude is in the product — and in how we build it.

Claude is the agent we build with and build for: inside TripPocket and ACP Gateway, and across our engineering workflow.

In TripPocket

Claude, connected over MCP.

Claude connects to TripPocket over MCP to search, cite and save travel knowledge on the user’s behalf, under scoped, revocable tokens.
In ACP Gateway

Claude Code, over the Agent Client Protocol.

ACP Gateway runs Claude Code via ACP, with session forking, model and thinking controls, and human approval of tool calls.
How we build

Agentic engineering, reviewed.

Claude Code is our main engineering workflow. Agents open issues and pull requests, run CI, and go through review before anything ships.

Independent, founder-led, early.

GECHEN AI is an early-stage startup in Taipei. We build AI that helps with real life, on your terms, and we keep it open to any agent through MCP and open APIs.

Founder: George ChenSoftware & AI systems engineer
Location: Taipei, Taiwan
Building since: 2026
Stage: Early-stage, bootstrapped, pre-revenue
Contact: george@gechen.org
GitHub: github.com/bamoo456

Try the products. Read the code.

TripPocket is live in beta and ACP Gateway is open source under MIT. For anything else, write to george@gechen.org.

Open TripPocket (opens in a new tab) ACP Gateway on GitHub (opens in a new tab)

AI that helps with real life, on your terms.

>What we’re building: software agents can use.

Two interfaces, one domain model

Scoped Agent Access Tokens

Provenance on every item

See the work, not just the chat

Approve from web or phone

Branch or hand off

A native iOS console

>Claude is in the product — and in how we build it.

In TripPocket

In ACP Gateway

How we build

>Independent, founder-led, early.

>Try the products. Read the code.

What we’re building: software agents can use.

Claude is in the product — and in how we build it.

Independent, founder-led, early.

Try the products. Read the code.