The most interesting AI software being built right now isn't the model. It's the loop around the model. validate, lock, run, compress, repeat. Liminal is that loop, made visible, owned, and yours.
What we're shipping
Liminal AI is a model-agnostic agent harness. Everything you need to run a real agent on real software work:
- A ReAct loop with retries, provider circuit breakers, context compression, drift scoring, resumable streaming writes, and approval gates on destructive tools.
- Lazy-loaded tool families. files, shell, git, code intelligence (semantic rename), web, browser, Obsidian vault, document engine (PPTX/DOCX/PDF), and more.
- Memory that compounds. typed notes, hybrid BM25 + vector recall, recipe library for successful multi-tool turns, optional background auto-dream consolidation.
- A persona system. voice, behaviors, and a themed web shell from a single bootstrap prompt.
- Two UIs. Ink terminal for keyboard-driven work; Express + React web UI streaming tool calls over SSE.
Why a harness, not a hosted agent
Three things bother us about hosted agent products:
- They're opaque. you can't see why a tool fired or what's in memory.
- They lock the model. you can't swap DeepSeek for Claude on the same workload.
- They take margin on spend. every token is marked up through their layer.
Liminal is the inverse: open source, local-first, JSONL session traces, MIT licensed.
The trade
You run it yourself. Node.js 22+, a provider key, a few minutes of setup. In exchange you get a serious harness with infrastructure that survives the failure modes that kill most demos.
How to try it
Install (macOS / Linux / WSL)
curl -fsSL https://www.vireondynamics.com/install/install.sh | bashInstall (Windows PowerShell)
irm https://www.vireondynamics.com/install/install.ps1 | iexAdd your API key
The wizard configures OpenRouter or another OpenAI-compatible provider.Open the UI
Terminal or web: same harness underneath.
Try Liminal on your machine
Free to use (FSL-1.1-MIT), one-command install. No account required.
We've been building this for months. v0.1 is the foundation: feedback welcome on GitHub or admin@vireondynamics.com. The Vireon Dynamics team