>_ context-transfer download for mac

$ pbpaste | context-transfer --extract

Your context shouldn't
live and die in one chat window.

Select text anywhere, hit a shortcut, get a structured card that works in any AI tool. Runs fully local.

download for mac view on github
context-transfer --capture 
$ pbpaste | context-transfer --extract
// 3,847 chars of raw chat, pasted in
me: ok so the migration is blocked until
the mobile team signs off on refresh...
✓ extracted in 0.8s · ollama · 0 network calls
┌ CONTEXT CARD ────────────────────────┐
Goal          ship JWT migration without breaking mobile
Decisions     httpOnly refresh, 15min access tokens
State         middleware passing tests, PR #142 open
Open          device-bound refresh tokens?
└────────────────────────── copied ✓ ───┘
capture pressed anywhere → structured card, in about a second press E to replay

# the problem

Your context is trapped.

Every AI tool keeps its context in its own silo. Start a thread in one app, move to another, and you're re-explaining your project, your constraints, and your decisions from scratch.

The context wasn't lost. It just can't leave.

ChatGPTcontext locked in session
Claudecontext locked in session
local llamacontext locked in session
new chat, same youcontext: 0 tokens

Three moves. No dashboard.

  1. Select text anywhere

    Highlight a conversation in any app, or paste a raw transcript straight into the panel. If you can select it, you can capture it.

    anything with selectable text
    ⌘⇧E works system-wide on macOS

  2. Hit extract

    One shortcut or one button. A local model reads the mess and pulls out what actually matters.

    Context Transfer
    ✓ extracting via llama3.1:8b

  3. Copy the card

    Goal, decisions, constraints, current state. One Copy button and the context travels with you.

# local-first, by default

Nothing leaves the machine.

Point it at Ollama and extraction runs on your own hardware. No API key, no cloud call, no telemetry. This is not another wrapper around someone's API.

  • No API key required
  • No network calls on the extraction path
  • No telemetry, no accounts, no cloud
  • Open source, MIT licensed
wifi: connected
ollama ps 
$ ollama ps
NAME           ID         SIZE     PROCESSOR
llama3.1:8b    4f3a9d2e   5.2 GB   100% GPU

$ context-transfer --capture
“extract the decisions and constraints...”

✓ extraction complete in 0.8s
✓ 0 network requests made

# what gets extracted

One card, carried anywhere.

A messy 4,000-token conversation goes in. This comes out. Copy it into any chat, any tool, any model.

Context Card ✓ copied

Captured on

Sep 14, 2026, 11:42 PM

Goal

Migrate the auth service from session cookies to short-lived JWTs without breaking the mobile clients.

Key Decisions

  • Refresh tokens live in httpOnly cookies
  • 15-minute access tokens
  • Skip token rotation for v1

Constraints & Preferences

  • iOS client ships weekly; no breaking API changes before Oct 1
  • Prefer boring, well-documented libraries

Current State

Middleware rewritten and passing tests; the mobile team is reviewing the token refresh PR.

Resources & Links

  • PR #142: token refresh flow
  • ADR-019: sessions to JWT

Open Questions

  • Do we need device-bound refresh tokens?
  • Who owns the logout-everywhere endpoint?

Real output, real format. The Copy button on your extracted cards works exactly like this one.

# bring your own model

Local by default, cloud when you want it.

$ context-transfer --backend ollama

Runs on your machine via Ollama. Private by default, works offline, and the card format is identical regardless of model.

# faq

Questions, answered.

Does Context Transfer work with ChatGPT and Claude?

Yes — with any app that has selectable text. Highlight a conversation in ChatGPT, Claude, Gemini, a local model, anything. Context Transfer distills it into one structured card that pastes cleanly into any AI tool, IDE, or doc.

Is my conversation data sent to a server?

No. With the default Ollama backend, extraction runs entirely on your machine — no API key, no accounts, no telemetry, zero network calls on the extraction path. A cloud backend (Anthropic, NVIDIA NIM) is available, but only when you explicitly select it.

Why not just copy-paste the whole chat?

A raw transcript is noisy: introductions, wrong turns, tangents. The card distills a 4,000-token conversation into its goal, decisions, constraints, and open questions — smaller, faster to re-ingest, and consistent across models.

What does it cost?

Nothing. Context Transfer is MIT-licensed open source, and local extraction through Ollama is free forever — your hardware, your model, your data.

What macOS version is required?

macOS 14 or later, plus Ollama with a pulled model such as llama3.1:8b for local extraction. To build from source you'll also need Xcode 15+.

More questions? open an issue or read the README.

Start carrying your context.

Free, open source, MIT licensed. Download the .dmg, or compile it yourself in two minutes.

download for mac build from source

The .dmg is ad-hoc signed, not notarized — right-click the app and choose Open on first launch to skip the Gatekeeper warning.

zsh 
$ git clone https://github.com/Shivala-08/context-shifter.git
$ cd context-transfer && open ContextTransfer.xcodeproj
$ xcodebuild -scheme ContextTransfer -configuration Release build
✓ build succeeded · run the app, press ⌘⇧E
github.com/Shivala-08/context-shifter macOS 14+ Ollama for local extraction tip: press ⌘K anywhere on this page