When intelligence is the commodity, knowledge is the advantage.
Radarist is an open-source, local-first innovation platform that turns weak signals into decisions. It runs on your machine, and the knowledge it builds is yours.
The shift
Intelligence is getting cheaper. Knowledge isn’t.
Frontier models are converging. Each release narrows the gap, and the cost of running them keeps falling — roughly tenfold a year. Raw intelligence is becoming something you rent by the token.
When everyone can buy the same intelligence, it stops being an edge. The advantage moves to what the intelligence is pointed at: your data, your history, your decisions. The things no competitor can download.
That advantage isn’t bought. It accumulates. The model is the engine. Your knowledge is the fuel — and the map.
evidence-backed knowledge graph — yours, compounding
How it works
A single loop that compounds.
Radarist turns weak signals into decisions through one repeating loop. Each pass adds evidence to a knowledge graph that grows on your machine and gets sharper every time you use it.
- Signaldiscover & triage
- Entityresolve
- Relationlink with evidence
- Radarplace & track
- Reportdecide
The principle
Own your advantage.
Knowledge is only an advantage if it stays yours. So Radarist is built around four commitments.
Run on local
Start fully on your machine. No account, no upload, no waiting.
Grow on local
Your knowledge graph compounds where it lives, under your control.
Own your data
Yours by default. Export it, delete it, move it. No lock-in.
SaaS on demand
Hosting is opt-in, never required. You choose if and when to leave local.
Run on local. Grow on local. Own your data. SaaS on demand.
Where it’s going
Autonomy you can trust, because it’s earned.
Most agents claim autonomy on day one. Radarist makes them earn it. Every time you triage a signal, your decision becomes a label — a quiet record of what good judgment looks like. Over time those labels measure how reliable an agent actually is, not how confident it sounds.
An agent only graduates to acting on its own when its measured reliability clears a bar, set per relation-type and confidence band. Trust is calibrated, not asserted.
Graduate slow, demote fast. Earned trust rises one rung at a time; a slip drops it immediately. That is the difference between an agent that demos well and one you would actually rely on.
- Manualproven
You decide. The agent watches and learns.
- Assistedproven
The agent suggests. You approve every call.
measured reliabilitygate cleared - Supervisednext
The agent acts. You review by exception.
measured reliabilitybelow the bar - Autonomousthe horizon
The agent acts alone, within proven limits.
measured reliabilitybelow the bar
For builders
Clone it. Run it tonight.
Radarist is open source and runs locally. Clone the repo, point it at what you’re tracking, and watch the graph start to form. This is v0.1.0 — an early prototype with rough edges and a roadmap worth shaping.
$ git clone github.com/radarist/radarist