Pick up where
the conversation left off.

A new chat can lose track of earlier decisions and why you made them. You can't easily see what the AI remembers.

GFT Map turns your discussions into context you can inspect, correct, and carry into your next task.

Open source · Local storage · No GFT account required

GFT Map document and map interfaceOpen the editable English example ↗

Start with something familiar.

Three editable examples. No installation needed.

Product decision map
Product decisions

Why isn't bank sync in the first release?

Keep the scope, trade-offs, and next experiment together.

Explore the decisions ↗
Writing example
Writing

Keep discarded arguments out of the next draft.

Carry the author's position and unfinished ideas into your next writing session.

Edit the document ↗
Technical planning example
Technical planning

Keep the constraints when you start coding.

Preserve the recovery flow, idempotency requirements, and deferred changes.

Explore the plan ↗

Authored examples using the real app components, not user conversations or model evaluations. You can edit, inspect sources, and import or export. The demo does not call a model or access your chats.

One task, across conversations.

Collect

Connect a conversation and update your map.

Read the current understanding in Doc. Explore decisions and their relationships in Map. Received sources stay in Log.

Correct

Make the understanding yours.

Edit the topic, document, or nodes. Tidy when needed; redraw from saved sources when your scope changes.

Continue

Ask your next agent to read it.

The agent checks an index, then reads the content it needs. Your corrections are available on its next read; changes do not automatically wake the chat.

Take your context with you.

Copy the document, or download a complete map containing the topic, Doc, Map, and Log. Import it into your local app to keep working.

Try the round trip ↗
Download an English example ↓

Use your own conversations.

Install the Skill, ask your agent to open GFT Map, and connect a conversation.

1. Install

npx skills add CoralLips/gft-map --skill gft-map --agent codex --global

The installer requires Node.js 22.20+. Model actions require a configured agent executor.

2. Ask your agent

Use gft-map. Check the local environment, start the panel with an available executor, and give me its address. Help me connect a conversation and create a context map.

Codex execution has been tested with a real task. Reopen your session after installing the Skill if needed.

Installation and usage

A few practical details.

Does the demo run an agent?

No. Reading previews show the content an agent could read, not a generated response. Update, Tidy, and Redraw require the local app.

Does Claude Code work?

Chat sources are supported. Claude ACP execution and cross-client continuation remain experimental and require valid local authentication.

Can I connect to GFT?

Yes. Signing in enables automatic two-way map sync, including deletions. Without signing in, maps stay local. Model actions send the necessary material to your configured model service.