Know exactly what internet you are getting — and prove it.
Helix runs on your computer and measures your connection without stopping: latency, jitter, loss, outages, speed. When something breaks it tells you where — your Wi-Fi, your router, your ISP, or the far end — and hands you a report you can send to someone who can fix it.
Free tier, no card. Windows 10/11 on x64. The v1.0.0 installer is not code-signed yet — we explain exactly what you will see.
Your ISP measures its network. Nobody measures your line.
ISP-side quality tools are deployed and owned by the operator. Consumer tools are owned by you but tell the ISP nothing. The call ends with “we see no problem on our side” and no way to settle it. Helix is a customer-owned agent whose data the customer can consensually share with their ISP — so the conversation starts from the same evidence.
If you just want it to work
A 0–100 score and a plain-language answer to “is it my equipment, my line, or the wider internet?” Outages are logged while you are away from the desk, so the evening the call kept dropping is still on the record tomorrow.
If you do this for a living
Continuous multi-target probing, MTR with per-hop timelines and ASN annotation, loss classification that knows the difference between an ICMP-rate-limited router and real loss, and exportable evidence in JSON, CSV and WinMTR format.
If you run the network
A per-subscriber quality view sourced only from customers who opted in, tickets that arrive with the pre-fault history already attached, and an integration contract your NMS can generate types from. The ISP platform →
Four decisions that are unusual, and that we can show you.
Measurement products are easy to make look confident and hard to make honest. These are the four places Helix chose honest, and each one is visible in the product rather than only on this page.
Counts on the wire, never percentages
A percentage cannot be re-aggregated: average four “2 %” windows and you get a number that means nothing. Helix uploads loss counts, the longest consecutive-loss run, and where runs cross a window boundary, so any span you ask for later is recomputed exactly rather than averaged badly.
Below 300 probes, a loss figure is labelled coarse instead of quoted as a rate — because at that sample size it cannot honestly distinguish “fine” from “a bit lossy”.
A verdict has to state what argues against it
Every diagnosis Helix produces carries two lists: supporting evidence and contradicting evidence. Both are always rendered. A verdict engine that cannot populate the second one does not ship.
That rule is what stops the classic false accusation — per-hop loss on a middle router that does not persist to the final hop is a router deprioritising ICMP replies to itself, not a broken path, and the report says so.
Sharing is off, and revocation bites immediately
Your ISP sees nothing until you turn sharing on, and you choose which connection. Turn it off and the next query the ISP runs returns nothing — there is no cache to expire and no nightly job to wait for.
Two independent gates decide it: consent (did you say yes, and is it still yes) and scope (which of your connections this particular ISP may see). Both fail closed. How that works →
The ISP integration contract is public
The schema an ISP's network-management system integrates against is served unauthenticated, so anyone can read what is exchanged and generate types from it rather than trusting a description.
It is deliberately never stricter than the server: where Helix tolerates a value instead of refusing it, the schema says so. Read the schema →
Install, measure, diagnose, and only then share.
-
Install the agent
A Windows service does the measuring and a desktop app views it. The split matters: monitoring continues whether or not the window is open, and the app itself never runs elevated.
-
It measures continuously
Multi-target probing at one packet per second, DNS timing every minute, bandwidth accounting, scheduled speed tests with a bufferbloat grade. Everything is stored locally first, so the agent is useful before you have an account at all.
-
Run the Link Troubleshooter when something is wrong
A twelve-stage examination from your network adapter to the destination — adapter, Wi-Fi radio, gateway, addressing and CGNAT, DNS, captive portal, identity, reachability, path analysis, MTU, throughput, and the last 24 hours of history. Out comes a verdict with its confidence and its counter-evidence, as a report you can save, publish at a revocable link, or attach to a ticket.
-
Share it — if you want to
Sign in to the cloud dashboard for history across devices, then optionally link your ISP account and turn on sharing. From there a support ticket arrives at your ISP already carrying the evidence, tagged with the service ID they know you by.
Modes change the vocabulary, never the permissions.
The same measurements are worth different things to different people, so the dashboard has four modes that reorder the panels, change the density and phrase results for the reader. A mode is presentation only — it never grants or withholds a capability, and every panel is present in every mode.
Home
Is it my equipment, my line, or the wider internet?
Gamer / Streamer
Jitter and loss runs first — the numbers that ruin a match.
Enterprise
Evidence you can forward: windows named, nothing invented.
Operator
Every number, no rounding and no whitespace.
Start measuring your own line.
The free tier keeps three days of readable history and costs nothing. The agent runs locally either way.