Terms & Privacy
Taking part in the ideathon
Two things live on this page: the terms you accept by taking part, and exactly what the Ideathon MCP server records when you use it. The second half is written to be checkable — every claim on it corresponds to something in the code, and where a limit exists it is stated rather than smoothed over.
1 · Taking part
By submitting an idea you accept the official rules of the ideathon, which cover eligibility, what counts as a valid submission, the per-track submission windows, how ideas are judged against the published rubric, prizes and payout, originality and ownership, and disqualification.
Those rules are maintained in full, in public, here: the official Ideathon rules ↗. They are the binding version — if anything on this page and the rules disagree, the rules win.
You keep ownership of your idea. Submitting it does not transfer it to Arkiv or to Golem.
2 · What the MCP server records
The Ideathon MCP server is the read-only knowledge base your agent can connect to. It needs no account and no key. It records a small amount of information about each interaction, for one reason: to find the questions the documentation fails to answer, and fix them while the event is still running.
Never recorded
There is no account and no sign-up, so we never ask you for your name, your email address or any other contact detail, and there is no field anywhere to put one in. What follows is never recorded either:
- IP addresses
- Your location or country
- Browser, device or User-Agent information
- Cookies, session identifiers or device identifiers
- Any key that links two requests to the same person
- The text of your idea — not the text, not a summary, not a hash of it
Recorded, per interaction
- The time, which tool was called, whether it succeeded, and how long it took.
-
Which MCP client family you connected with, reduced to a fixed list (for example
claude-codeorcursor) plus its major version number. The raw client name your software sends is not kept. - For a document request: which document, from the fixed list of documents in the knowledge base, and whether it existed. Your AI assistant may also attach the question that prompted the request; when it does, that question is stored under exactly the same masking and length rules as a search (next item).
- For a search: your search terms, with anything shaped like an email address, a URL, an IP address, a hex string or a long run of digits replaced by a placeholder before storage. Searches longer than twelve words are not stored at all. We also record how many results it returned and whether it returned none.
- For the idea coach: only a broad size range of what you pasted (for example “200–800 characters”) and the track you declared. Never the idea.
Said precisely
On IP addresses. No IP address is ever written to our telemetry database. The server does read the address of an incoming request in memory, to enforce a rate limit, and our hosting provider sees it in order to route the request and in short-lived security logs. We are telling you that nothing is stored — not that no server ever sees it.
On search terms and attached questions — the free text we keep. A search (or the question your assistant attaches to a document request) is free text, so it is the only place where you could put personal information into this system. The masking above removes things that look like identifiers; it cannot recognise a name. If you type a name into a search, that is what gets stored. So: please keep searches to keywords, and do not put personal or confidential information in them — yours or anyone else’s.
On counting. Because nothing links two requests together, this data counts calls, not people. We cannot tell how many individuals used the server, and we will not present a figure that claims otherwise.
On the landing page. ideathon.arkiv.network loads a 1×1 image that counts page views. What is stored is the page’s path and a timestamp — nothing else. No cookie is set, no IP address or browser fingerprint is stored, and nothing links two views together: it counts views, not visitors.
How long, and what it is not used for
Individual records are kept until 30 November 2026 — the event closes on 31 August and the write-up follows it. After that date only anonymous aggregate counts remain. This data is not used for advertising, profiling or automated decisions about anyone, it is not sold or shared for those purposes, and it is not linked to your submission. Your submission itself is handled separately through the submission form.
Contact
Questions, objections, or a request to remove something: write to privacy@golem.network. Because the telemetry holds no identifier, we usually cannot tell which rows relate to you, and we will not collect extra information about you in order to find out. If you want something specific removed, include details such as the approximate time and the search text and we will look for it.