About
We accept telemetry and keep it. That is the entire product.
Halkon is a telemetry ingest and retention service. It does not do application performance management, business intelligence, incident management, or artificial intelligence. It accepts what your servers send, stores it where you said to store it, and lets you read it back.
The company was founded in 2019 by people who had spent the previous decade operating observability stacks for other companies and were tired of two specific things: invoices that could not be predicted from the fleet, and vendors who could not say in one sentence which country a log line was sitting in.
Operating rules we hold ourselves to
These are the promises that shaped the architecture, written as constraints rather than values. Each one costs us something, which is how you can tell we mean them.
The price list is public and complete
Every number that can appear on an invoice is on the pricing page. There is no enterprise tier with hidden arithmetic and no quote that depends on how much we think you can pay. Price changes take 90 days’ written notice.
Data does not cross a region boundary
Not for backup, not for indexing, not for support, not for a migration we found convenient. The itemised list of what does cross has two rows on it, and both are counters rather than content.
We refuse loudly rather than bill quietly
Exceeding a ceiling produces 503 and shedding in a documented order, not an overage line. A runaway debug flag should wake someone up, not arrive as a surprise at the end of the month.
Nothing is dropped silently
Truncation is flagged on the event. Shedding returns a code and a class. Rejection names the rule that was broken and the value that broke it. If you cannot find out what happened to a byte, that is a defect.
Open formats on the wire, both directions
We accept OTLP, Prometheus remote write, syslog and plain HTTP. We emit NDJSON, CSV, Arrow and a scrapeable text format. There is nothing proprietary between your producer and your dashboard, and leaving does not require our cooperation.
Incidents stay published
The incident log is append-only. Entries are not removed when they age or when they are embarrassing, and every one names what changed afterwards.
Build history
| Date | Age | Event |
|---|---|---|
| 2019-04 | — | Company founded. Two people, one prototype ingest process, no storage layer. |
| 2020-08 | — | Private beta on a single machine. Eleven accounts, 400 GB accepted in the first month. |
| 2021-03 | — | eu-north-1 becomes the first production ingest point. Agent 1.0. |
| 2022-01 | — | na-east-1 live. The region model — one account, one jurisdiction — is fixed here and has not changed since. |
| 2022-06 | — | eu-central-1 live. Disk buffering ships in agent 2.0 after a customer outage lost four hours of logs. |
| 2023-02 | — | eu-west-1 live. Columnar block format v1 replaces the row store; storage cost per accepted GB falls by 61%. |
| 2023-11 | — | Published price list replaces per-account quotes. Three plans, no negotiation on the per-GB rate. |
| 2024-04 | — | ap-southeast-1 live. First region opened on legal demand rather than latency demand. |
| 2025-09 | — | sa-east-1 live. Six ingest points, 4.75 M ev/s combined ceiling. |
| 2026-05-19 | — | HKQL v3. Cost classes become visible before a query runs; join is removed in favour of bounded lookup. |
| 2026-08-19 | — | Platform peak of 3.10 M accepted events/s, and the partial outage that came with it. |
| 2026-09-08 | — | Agent 4.2.1, the current release. |
Dates are fixed. The age column is computed in your browser from these dates and your current clock.
How the company is arranged
Eleven reliability engineers carry the pager across all six regions on a follow-the-sun rotation, and every engineer who writes ingest or storage code takes a share of it. There is no separate operations department to hand a broken release to.
Support is staffed by engineers, not by a scripted first line. A Regional P1 reaches a person who can read the ingest fleet’s own telemetry within the hour, at any hour.
We take no advertising, run no analytics on this site, and have no growth team. The product is sold by being cheaper to predict than the alternatives and by publishing enough detail that an engineer can decide without a call.
Our on-call engineers are reachable through the support channel, under an account, with an audit trail. Publishing individuals as marketing surface is how people end up receiving out-of-band pressure during an incident, and we would rather they did not.
Contact
Pick the channel that matches the question. Technical questions answered by engineers; commercial questions answered with arithmetic.
Or write here
This form does not transmit anything. It is intercepted in your browser and shown back to you — this site is a static build with no server side and no third-party script of any kind.