Sometimes building YveChat is two people asking the same question from completely different directions. 😄

Dan thinks ownership. Alvis thinks infrastructure. They debate, challenge each other and occasionally create more work for Alvis.

Meet the brains behind YveChat. https://yvechat.xyz/news/the-conversations-behind-yvechat

In just 5 days, Total Engines grew 66%.

Engines alive: 382 → 629
Daily active Engines: 241 → 400
Daily inferences: 4,850 → 9,039
Attesting wallets: 208 → 270

More people putting their own machines to work. Humbled by the growth!

Dashboard: yvechat.xyz/#achievements
On-chain Transparency: https://yvechat.gitbook.io/wp/res/method

Six digits connect the web workspace to your machine.

YveChat uses a one-time pairing flow between the hosted Workspace and the local Engine. You approve the introduction yourself on your machine. Pairing includes origin binding, hashed storage, expiry and revocation. The origin policy accepts configured origins and refuses wildcards.

Once paired, the browser has a controlled path to the Engine operating locally. A small interaction establishes a very important trust boundary.

Local AI is getting seriously capable. Its infrastructure is the next frontier. ⚙️

CTO Alvis Sim breaks down Ollama, self hosted data, model choice, agent permissions and why YveChat Engine exists.

The models are moving fast. Here is what needs to catch up https://yvechat.xyz/news/local-ai-needs-better-infrastructure

Longer context has a physical cost.

With Ollama, memory demand scales with parallelism and context. More simultaneous workloads and larger context windows require more resources.

Running intelligence yourself means seeing the real hardware economics underneath it.

Local AI can grow beyond a single desktop runtime.

YveChat's experimental OpenAI-compatible adapter supports vLLM for users with suitable GPU infrastructure, extending the same workspace toward independently operated model-serving environments.

Your machine already has an AI environment. YveChat Engine discovers it.

Dynamic model discovery identifies available local runtimes and models, turning infrastructure running on your own hardware into something the workspace can understand and use directly.

A wallet signature has a clear boundary.

SIWE authenticates control of a wallet. Phishing resistance depends on wallet security, wallet-derived encryption lacks a standard, and signature identity has no built-in recovery.

Sovereignty works best when its limits are explicit.

Sovereignty covers the entire AI workspace.

Compute. Data. Access. Identity. YveChat places each layer within the user's control, from local runtimes and storage to pairing and optional wallet signatures.

Read the full architecture https://yvechat.gitbook.io/wp

Sovereign inference has a real hardware layer.

Local GPU acceleration can be driver-sensitive. vLLM requires suitable GPU infrastructure.

Hardware, drivers, runtimes, and context limits become visible parts of the stack when you operate the intelligence yourself.

Every model brings its own capabilities.

YveChat uses capability filters so the workspace can reflect what each connected runtime actually supports.

Provider neutrality means treating runtimes as peers while respecting the technical reality of each model.

The destination is simple: your thinking belongs to you.

YveChat is building toward a world where the record of how a person thinks stays under that person's control, even when sovereignty carries setup and operational responsibility.

Follow the build with us! https://x.com/yvechat

Web3 already established the custody instinct.

Hold the keys. Verify ownership. Run the infrastructure.

YveChat extends that same principle into AI, where months of prompts, research, drafts, and decisions can become a remarkably detailed record of how someone thinks.

Reading the web still carries a security boundary.

URL-reading tools can become SSRF primitives.

YveChat currently labels open_url, find_on_page, extract_article, and read_pdf Experimental while these capabilities and their security boundaries continue to mature.

The website is intentionally thin.

It renders the YveChat Workspace: threads, model selection, tool activity and citations. For local inference, Ollama runs the model on your hardware and SQLite holds the conversation on your disk.

The hosted interface becomes a window into infrastructure you control.

Four layers make the architecture legible.

Workspace renders the interface. Engine manages secrets, normalization, and execution. Runtimes perform inference. Storage persists conversations on user-controlled infrastructure.

Every layer has a defined boundary.

Intelligence stays outside YveChat's margin. Permanently.

Sovereign Pass is designed around software value including extended agents, team workspaces, encrypted sync, priority builds, and deployment support.

Users pay providers directly or run models on their own hardware.

Shipping means someone outside the team can actually use it.

Each YveChat roadmap era completes when its deliverables work on a user's own hardware without assistance. Progress is measured through paired Engines, locally served models, and threads persisted to user infrastructure.

https://app.yvechat.xyz/

Bring the endpoint. Keep the workspace.

YveChat experimentally supports manually configured OpenAI-compatible endpoints through a custom base URL, opening the workspace to compatible infrastructure beyond predefined runtimes.

Explore YveChat https://yvechat.xyz/

AI access moved fast. Custody is the next question.

Models became easy to reach while accounts, infrastructure, policies, and accumulated work remained provider-controlled.

YveChat starts from a deeper question: who ultimately holds the intelligence workspace?