kolm  /  community  /  discord server bootstrap

Discord server bootstrap.

Internal doc for setting up and running the kolm Discord. Five channels, three roles, light-touch moderation, no bots day one. Read top-to-bottom before opening invites.

Invite.

Placeholder URL: https://discord.gg/kolm. The vanity URL is reserved but the invite is gated for the first 90 days. Membership is "DM founders@kolm.ai with your GitHub handle or your work email." We approve manually until we hit 200 members; after that we open the gate.

The gating is not gatekeeping. It is a way to keep the early signal-to-noise high enough that the first 50 conversations set the tone of the server.

Server structure.

Five channels. No more, no fewer, for the first six months. A bigger channel list signals a bigger server; we are not that yet and pretending otherwise wastes attention.

  • #welcome. Pinned: read-only intro, link to /docs, link to /spec/rs-1, link to the CoC. New-member rule: post a one-liner about what you are building. No introductions = no other channel access (manual). This is the only read-only channel for non-staff.
  • #announcements. Read-only for everyone except staff. New release notes, breaking changes, security advisories, the HN/X launch links. Posts are signed by a real human, not a bot.
  • #help. The default channel for "I am stuck." Format hint in the topic: paste the recipe, paste the error, paste the kolm version. Threads opened liberally so long topics do not bury the channel.
  • #showcase. What you compiled. CID welcomed; no need to share the artifact itself. Brag a little; that is the point of the channel. Staff occasionally reposts highlights to /showcase on the public site (with permission).
  • #compliance-q-and-a. The narrowest channel by design. HIPAA, SR 11-7, NIST AI RMF, EU AI Act questions. Compliance officers and security engineers. Tone is "I am not your lawyer," because we are not.

Roles.

Three tiers. Permissions are deliberately flat.

  • member. Default for everyone who passes the welcome step. Read + write in #help, #showcase, #compliance-q-and-a. React-add in #announcements.
  • contributor. Anyone with a merged PR on the public repo. Auto-granted by the verify-on-PR Action once we wire it (manual until then). Same channels as member plus the ability to start polls and pin messages in #showcase.
  • staff. Founders and core maintainers. Send messages in #welcome and #announcements, kick/ban, edit pinned messages. Names visible. No anonymous moderation.

Pinned message templates.

The first pinned message in each channel sets the contract for the channel. Copy verbatim.

# #welcome
Hi - this is the kolm Discord. Read these three things first:

1. /docs is the canonical reference. Start there.
2. /spec/rs-1 is the artifact spec. Read it before you ask about manifests.
3. Code of Conduct: Contributor Covenant 2.1. See CODE_OF_CONDUCT.md in the repo.

Then post a one-liner about what you are building. We open the rest of the server once you have introduced yourself.

Direct line: founders@kolm.ai

# #help
Three lines that help us help you:
1. Your kolm version: `kolm --version`
2. The recipe you are running (paste it, not a screenshot).
3. The exact error or the CID of the artifact that fails.

For long async threads, GitHub Discussions is the better surface:
github.com/sneaky-hippo/kolmogorov-stack/discussions

# #compliance-q-and-a
We are engineers, not your lawyers. Answers here are best-effort engineering interpretations of public frameworks. Use them as a starting point, not as legal advice. /compliance is the canonical mapping.

Sensitive disclosures (potential PHI exposure, customer-data incidents) go to founders@kolm.ai directly. Do not paste private data here.

Moderation philosophy.

Light touch. Three rules.

  • No spam. One warning, then ban. No tolerance. Crypto airdrops, "earn $5k a week," DM-trading lists: instant ban, no warning.
  • Disagreements stay technical. If the conversation shifts from "your number is wrong" to "you are wrong," a staff member steps in. We use Contributor Covenant 2.1 wording for the warning.
  • Point long async to GitHub Discussions. Discord is great for "here is my error, help me debug." It is terrible at "what should the v0.2 receipt shape be." For the second, we move the thread to GitHub Discussions and link it back from Discord.

Public site moderation log lives at github.com/sneaky-hippo/kolmogorov-stack/blob/main/COMMUNITY_LOG.md. Bans are recorded with reason. Names are redacted to handles. Receipts on moderation are the same principle as receipts on AI.

Bot config notes.

None initially. We have seen too many Discord servers ruined by 14 bots that auto-respond to every keyword. The signal-to-noise drops the day the first bot lands.

We may add one bot later: kolm-verify-bot. Scope is narrow. A user pastes a CID in any channel and the bot replies with a lookup against /leaderboard: "CID found, base = Qwen3-32B, K-score = 0.974, last seen 2 days ago" or "CID not in public registry." That bot is on the v0.2 backlog, not v0.1.

No mod bots. Staff handles moderation manually for as long as the server fits in one human's head. Past that, we will reconsider, but not before.

Day-one checklist.

  • 1. Create the server, set the icon (kolm logo), set the boost-tier description.
  • 2. Create the 5 channels in the order above.
  • 3. Set up 3 roles with the permissions described above.
  • 4. Paste the 3 pinned message templates.
  • 5. Create the vanity URL discord.gg/kolm and update the placeholder on /community.
  • 6. Send the first 10 invites manually. Track the names in a spreadsheet so we can ask follow-up questions in the first week.