The Principle

Self-custody, now for your intelligence.

The principle that reshaped money applies to thought.

When the model runs on your machine, the workspace is yours to command.

YveChat carries self-custody into the way you work with AI.

session.yveyvechat
$
The Engine

Pair the workspace to your machine.

YveChat Engine runs on your computer and connects the hosted workspace to your local runtimes through an explicit pairing flow. A six-digit code links the two, and you approve it locally.

Available

Pairing flow

427193
  1. 1YveChat Engine runs on your computer.
  2. 2The workspace shows a six-digit code.
  3. 3You approve the code on your machine.
  4. 4The workspace reaches your local runtimes.

What stays on your machine

  • Models
  • Conversations
  • Generated images
  • Credentials
  • Execution

Why the Engine exists

01

Ollama origins

Ollama accepts browser requests from 127.0.0.1 alone by default. The Engine talks to Ollama locally, so you leave OLLAMA_ORIGINS untouched.

Read the source
02

Chrome Local Network Access

Chrome gates access from public sites to loopback and private ranges behind a permission prompt, now extended to WebSocket and WebTransport.

Read the source
03

WebKit loopback history

WebKit carried a loopback mixed-content issue since 2017. Safari 18 blocked HTTPS pages from loading http://127.0.0.1 until a March 2025 fix.

Read the source

An unsigned YveChat Engine developer-preview archive is available for Windows, macOS, and Linux. Signed production installers and automatic updates remain planned.

Hosted interface, local machine.

Four layers, each with its state. The hosted page pairs to the Engine on your computer, and your runtimes and data stay local.

// hosted page, paired to your machine

OpenAI-compatible endpoints and page reading are experimental. Hosted-provider key management, MCP, local document retrieval, and sandboxed code execution are planned.

L01

Hosted Interface

Available
app.yvechat.xyz

A static workspace served from app.yvechat.xyz. Models and conversations stay on your machine.

> interface = static.hosted

L02

YveChat Engine

Available
Your machine

A connector on your computer, paired by a six-digit code with your local approval.

> engine.pair = six.digit.code

L03

Local Runtimes

Available
Ollama, ComfyUI

Ollama serves language models and ComfyUI renders images, both on your hardware.

> runtime = ollama + comfyui

L04

Local Data

Available
Your custody

Conversations, images, credentials, and execution stay on your machine.

> data.location = your.machine

Capabilities

Every capability, with its status.

Available runs today. Experimental works with limits. Planned sits on the roadmap.

AvailableWorks today in the developer preview.ExperimentalPartially working, with limits.PlannedOn the roadmap for a future release.
01

Local language models

Ollama serves open models on your hardware. Prompts, output, and weights stay on your machine.

Available
02

YveChat Engine pairing

A six-digit code links the hosted workspace to your computer, and you approve it locally.

Available
03

Local image generation

ComfyUI renders images on your machine through image workflows discovered on your system.

Available
04

Web search with citations

Web search runs through the local tool gateway and returns clickable sources.

Available
05

Local and wallet sign-in

Sign in with a local account. Wallet authentication is available and optional.

Available
06

OpenAI-compatible and vLLM endpoints

Configured OpenAI-compatible and vLLM endpoints have limited support today.

Experimental
07

Page and PDF reading

open_url, find_on_page, extract_article and read_pdf run inside the local tool gateway. Chat orchestration uses search today, so these are reachable ahead of full integration.

Experimental
08

Hosted-provider key management

One OpenAI-compatible endpoint accepts a key today. Full management across the major providers sits on the roadmap.

Planned
09

MCP connections

Model Context Protocol connections sit on the roadmap.

Planned
10

Local document retrieval

Reading and retrieval over files on your own disk sits on the roadmap.

Planned
11

Sandboxed code execution

Running code in a local sandbox sits on the roadmap.

Planned
Verify it yourself

Proof you can run.

Four claims you can check for yourself.

pass

Local models answer with nothing leaving your machine.

Open your network inspector while a local model answers. Watch for zero outbound requests.

Verify
pass

The workspace pairs only after you approve a code.

Start the pairing flow. The workspace connects once you approve the six-digit code on your machine.

Verify
pass

A local model answers offline.

Disconnect from the network and run an Ollama-served model. It answers.

Verify
pass

Model availability matches what you installed.

Pull a model with Ollama, then reload. YveChat lists the models present on your machine.

Verify
Troubleshooting

When something stalls.

Six common states and the step that clears each.

YveChat Engine unavailable

Confirm the Engine is running on your computer, then pair again. Restart it if the workspace cannot reach it.

Browser local-network permission denied

Grant local-network access in the browser prompt, then reload the workspace.

Ollama unavailable

Start Ollama and confirm it is listening. The Engine reaches it once it is running.

No models detected

Download a model with Ollama, then refresh. YveChat discovers installed models after the Engine connects.

Pairing code expired

Request a new six-digit code and approve it on your machine.

Model too large for available hardware

Choose a smaller model. Larger models need more memory and stronger hardware.