LOCAL-FIRST AI

OLLAMA · VLLM · LLAMA.CPP · LM STUDIO

Your models.
Your machine.
Full power.

Source-available TUI coding assistant for Ollama, LM Studio, vLLM, and any OpenAI-compatible server. Your tokens never leave your hardware. No cloud routing, no subscription, no data you don't control.

Install uplnk
↓ — / wk · v1.1.0
View on GitHub
— · —
Works with ◆ Ollama ◆ vLLM ◆ LM Studio ◆ llama.cpp
Retro sci-fi transmission tower crackling with lightning in a dark star-filled void — the uplnk brand illustration

The situation

You've been here before.

01

It's 11pm. You're debugging a production issue. You paste your auth middleware into the chat window. The response is exactly what you needed. Then you remember: you just sent your session token schema to a vendor's inference server. You close the tab and figure it out yourself.

◆ With uplnk: same conversation. Same response quality. The inference runs on your machine. The tokens don't leave.

02

You installed Aider because it supports Ollama. It does — sort of. The streaming drops out halfway through a long response. The edit format confuses the 8B model. You spend 20 minutes reading error output instead of writing code. You paste the function into Jan.ai and keep going.

◆ With uplnk: the streaming layer handles local model latency. When a response stalls, you see why. When it succeeds, the output is syntax-highlighted and ready to use.

03

Your manager asks you to stop using Cursor for work. The security review found that source code is transmitted to third-party inference servers during normal use. You need an alternative that passes infosec review — and doesn't require explaining to the CISO what an ngrok tunnel is.

◆ With uplnk: local inference, no network egress, source-available code you can audit. The answer to your security team is two words: "runs locally."

Capabilities

Everything you need.
Nothing you don't.

TUI-Native

Runs inside your existing terminal. No browser, no Electron, no separate window stealing focus.

Local-First

Talk to Ollama, vLLM, LM Studio, or any OpenAI-compatible server. Your models, your rules, your hardware.

Smooth Streaming at ~30fps

Built for local model latency. A 33ms render-flush buffer keeps the UI smooth even when a hot local model pushes 60+ tokens/sec — tokens accumulate in a ref, not state, so React never thrashes.

Codebase RAG

Paste a file or enable local embeddings to index your whole project. uplnk chunks and stores vectors in SQLite — cosine-similarity search runs entirely on your machine. Opt-in — requires embedding endpoint

SQLite History

Every conversation saved locally at ~/.uplnk/db.sqlite. Portable, queryable, forever yours.

Zero Cloud

No API keys sent to our servers. No subscription required. Ever. Optional anonymous telemetry exists but is off by default and the transport isn't wired — nothing leaves your machine today.

Get started in 60 seconds

Three commands.
Running in under three minutes.

shell

# Install uplnk globally

$ brew install thepixelabs/tap/uplnk

# Connect your local LLM

$ uplnk config --provider ollama --url http://localhost:11434

# Start coding with AI

$ uplnk

◆ uplnk v0.1.0 — Connected: Ollama (llama3:8b) ●

Requires Node.js 20+ and a running Ollama/vLLM instance. Get Ollama →

Our commitments

What uplnk
doesn't do.

Features are easy to list. Refusing certain things takes intention. Here's what we've deliberately left out.

Phone home with your data

No telemetry, no analytics pipeline, no usage events. uplnk never opens a network connection you didn't initiate.

Require a subscription to function

Fair-code licensed. Install once, use forever. No account, no trial, no "free tier" that expires when you rely on it.

Force you into one model or vendor

If it speaks OpenAI-compatible API, uplnk talks to it. You pick the model, the hardware, and the server. We bring the tooling.

Open a browser window or Electron shell

uplnk is a TUI. It lives in your terminal, next to your editor, where you already work. No context-switch tax.

Hide the source code

Every line is on GitHub. You can read exactly what uplnk does before you run it — and fork it if we ever get it wrong.

Silently route your code to a cloud model

Cloud providers (Anthropic, OpenAI) are supported as explicit opt-in. Local inference is the default — not an afterthought bolted on later.

THESE ARE COMMITMENTS, NOT MARKETING. IF WE BREAK ONE, OPEN AN ISSUE.

Stop sending your codebase
to someone else's server.

Fair-code. PolyForm Shield licensed. No tracking. No paywalls. Install it once and it's yours — no subscription, no vendor lock-in, no account required.

Star on GitHub ★ — be first
↓ — downloads/wk · v1.1.0 · PolyForm Shield · Node ≥18