open source · local-first · free forever

The assistant that grows its own skills.

H.O.T-Jarvis writes and tests its own new abilities, remembers how it reasons, tells you when it's unsure, and lets you replay and undo anything it does — running on your machine, for nothing.

No account. No API bill. Nothing leaves your machine.

01

It grows its own skills.

Ask for a new ability and Jarvis writes the code and a test for it, proves the test passes, and refines on failure. Skills that can't prove themselves are flagged and refuse to run. The library compounds — every session, it can do a little more.

A glowing core emitting a skill card that self-tests and locks in with a check mark.
flagship · HyperFrames (HTML→MP4)
Points of light drawing faint constellations in dark space.
02

It remembers how it reasons.

Jarvis re-reads its own action log and keeps short lessons about what worked and what failed. Those lessons ride along in future prompts, so it improves from its own experience — not just from what you tell it.

03

It tells you when it's unsure.

Every answer carries the model's own confidence. Below a threshold, Jarvis asks a clarifying question instead of guessing. The gauge lives on the core, so you always know how much to trust the last thing it said.

A thin arc of light filling a circular gauge and settling.
A timeline of actions with one step reversing.
HyperFrames · authored animation
04

You can undo anything.

Every action is recorded with enough state to reverse it. Undo a chat turn, a saved note, or a skill change from the timeline — and run a replay audit that proves the log reproduces memory exactly. Nothing it does is permanent unless you want it to be.

It's yours. And it's free.

Local by default — inference runs on your own machine through Ollama, with free cloud tiers only as a fallback. No paid API is ever required, and your data never has to leave the room.

Apache-2.0 · Windows & macOS · built in the open