Matrix LLM · iPhone, iPad, Mac and Windows
Every answer shows the exact passage it is based on. Matrix also builds a knowledge graph of who relates to what, and every link in it has a quote from the source. The models can run entirely on your device: small ones on the iPhone, large ones on a Mac or PC.
Free to download for iPhone, iPad, Mac and Windows.
For people using Claude Code
A free plugin that gives Claude Code a local index of your PDFs, scans and charts. Every answer comes with file, page and passage, and it needs less than half the input tokens. It uses the same engine as the app.
Seconds per question
Input tokens per question
For people building with AI agents
It reads the support forums, writes the fix and has to pass checks that are plain code, so no model can argue its way past them. Every release, merge or forum reply then waits for a one-word OK from my phone.
How the loop worksFor WordPress site owners
A free plugin that stops form spam without tracking anyone. The visitor's browser solves a small computing task (a proof of work) instead. No cookies, no third-party calls, no puzzles to click. It runs on about 4,000 sites.
The software around the model mattered more than its size: a dense 9B with agentic search beat a 35B MoE with one-shot retrieval. Second finding: a 3B-active MoE kept up with a dense 27B.
How Matrix now sizes a local model's context window from free memory, why a cold start on the Mac got only 4,096 tokens, and what else moved this week.
Why a click next to a table froze a WebKit app for 7–12 s while accessibility was active, how a small selection guard fixed it, and what else moved this week.
