Local Agent Gateway

A lot of everyday work — transcribing a recording, tidying a spreadsheet, drafting a first pass at a document — gets routed through somebody else’s servers by default, even when nothing about the task requires it. That routing is a habit built into most software, not a technical necessity: the file goes up, the answer comes down, and in between it has sat on infrastructure the person doing the work has no visibility into and no control over. The same request often takes longer once it is queued behind everyone else sending similar requests to the same shared service, for work a single local machine could have finished on its own. For anyone handling something sensitive, working somewhere the connection isn’t reliable, or simply uncomfortable with that path for a task that doesn’t need it, the habit is the problem, not the AI.

These tools run entirely on the machine they’re installed on. There is no account to create and no network round trip for the work itself — a tool built for transcription, for instance, has been trialled running fully on-device rather than sending audio anywhere to be processed. Each tool keeps working without a network connection at all, which is not a fallback mode bolted on afterwards but the normal way they run. Models are bundled so that using a tool for the first time doesn’t require a separate download step before it works, which matters for exactly the situation the tools are built for: someone who needs it to work now, on the machine in front of them.

The set has been public since the start of 2026 and grows in the same way the rest of the portfolio does — a real task that kept costing more time than it should gets turned into a tool, and the tool is published once it’s clear the problem wasn’t unique to us.

Local AI Tools suits individuals and small teams doing hands-on work with sensitive material — documents, recordings, records — who want AI assistance without that material leaving their own machine. It suits people who already think carefully about where their files end up and would rather that judgement stayed in their own hands than be handled by a settings toggle they have to trust. It suits people working somewhere a reliable connection isn’t guaranteed, and anyone who would rather not depend on an account or a subscription to keep using a tool they rely on regularly. It’s aimed at one person at a time working on their own machine, not a shared team deployment.

It does not suit a team that wants a shared, centrally managed deployment across many machines with one admin console controlling everyone’s setup — these are tools for one person on their own machine, not infrastructure a team lead rolls out and monitors centrally. It does not suit workloads that need more computing power than the machine running them actually has: because nothing goes to somebody else’s servers, nothing else’s servers can help carry the load either. It also does not suit anyone who wants one update to reach every user identically and immediately — with no central deployment, updating an installed tool is a deliberate action on that machine, not something that happens invisibly overnight. And because the toolset is labelled Beta rather than a finished product, it does not suit someone who needs a guarantee that a specific tool will still exist, unchanged, next quarter — some of what’s here is still being trialled, and trials sometimes end.

Local AI Tools is free to use, with no separate paid tier attached to it. To start, get in touch through the contact page — DarkHorseOne’s engineering team reads every message and usually replies within two working days. There is no public download page yet: access starts with that message, the same way it does for the rest of the portfolio.