The community and support forum your AI can read.
Durable Q&A deflects repeat support and becomes the answer AI search engines cite. Bounded, auditable agents seed and tend the room, so a lean operator runs support without a team.
- human
- approval requiredapproval required
- 100%
- actions tracedactions traced
- 5
- grantable scopesgrantable scopes
- 7f3a…c1
- last tracelast trace
- 14:22agent:moderator-7community:writeproposed pinning a setup checklistapproved by MK
- 14:19rosa@acmecommunity:moderatemerged two duplicate topicsapplied
- 14:07agent:moderator-7community:readsummarised a thread for the digestread-only
One answer, resolved once, cited twice.
A member's question gets a durable answer on your domain. The same answer deflects the next person who asks and becomes the substrate an AI search engine cites when someone asks it instead.
Illustrative example, not a real thread or a real AI answer engine response.
Why did my scheduled export stop running overnight?
Scheduled exports pause automatically after three consecutive failures. Check Settings, then Exports, for the failure reason, fix it, then resume the schedule. This resolves the stall in under a minute.
"scheduled export stopped running overnight acme"
Scheduled exports on Acme pause after three failed runs in a row. Resolve the underlying error, then resume the schedule from Settings, then Exports.
acme.pubroo.link/supportThis is what the agent proposing something actually looks like.
One support thread, answered once. The answer stays readable on your domain, so it deflects the next person who asks and stays citable when an AI search engine goes looking for it. The agent that tidied it up still had to propose the change and wait for a human.
How should we welcome new members?
Proposed adding a pinned setup checklist to this topic.
An agent you can read like a contract.
- Granted
community:identityidentity onlycommunity:read2 tools listedcommunity:write3 tools listed- Not granted
community:moderate2 tools not listedcommunity:admin4 tools not listed
trace:7f3a…c1 · issued 14:02 GMT · revocable from your dashboard
Five scopes, three granted. The two it does not hold are not merely refused at call time: those tools are never listed to the agent at all.
- SCOPE
It can only do what you grant.
Permissions are explicit and minimal: community:read to search, community:write to propose, nothing implied. No standing access.
Read the scope reference - APPROVAL
Risky actions wait for you.
The agent proposes; a human approves. Destructive moves never auto-execute. They queue as a reviewable diff.
See the action lifecycle - AUDIT
Every action leaves a record.
Each decision is stamped with actor, scope, target, trace, and time. The ledger is immutable and keeps full audit ledger history.
Read a ledger entry
Horizontal, not a destination.
- PER PRODUCT
One layer, deployed per product.
Every product gets its own community and support forum at community.<productDomain>, not a shared destination.
- ISOLATED
Each community is its own database.
Data isolation is structural: a separate Durable Object SQLite database and its own Better Auth instance per community.
- ZERO CONFIG
Email auth works on day one.
Zero-config email authentication on creation. Social providers are opt-in advanced settings, never a setup requirement.
Look closer at any part of it.
Pricing that scales with the room.
| Plan | Monthly price | What you get | Action |
|---|---|---|---|
| Free | $0/mo | 1 community · acme.pubroo.link subdomain · logo and brand colour · propose-only Triage Agent | Start free |
| TeamPopular | $29/mo | 5 communities · custom domain · approve-gated agent · full audit history | Choose Team |
| Scale | $99/mo | unlimited communities · custom domains · admin MCP scopes · full audit | Choose Scale |
Bottom line: start free on a Pubroo subdomain, move to Team ($29/mo or $290/yr) for a custom domain and approval-gated agent, then choose Scale ($99/mo or $990/yr) for unlimited communities and admin scopes. Compare the plans in full.
How a scoped AI moderator works
As of .
What is an AI moderator with scoped permissions?
- It is a forum moderation agent that can only use the permissions you explicitly grant. Pubroo has five scopes: community:identity, community:read, community:write, community:moderate, and community:admin. There is no standing access and no wildcard. A tool outside the granted scope is not listed to the agent at all.
How does propose-and-approve moderation work?
- The agent never executes risky or destructive actions directly. It proposes a change as a reviewable diff, and a human with a sufficient role approves, edits, or rejects it. Only approved actions take effect, and rejections are recorded too.
What does Pubroo's audit ledger record?
- Every action is stamped with the actor, the scope it used, the target it touched, a trace identifier, and the time it happened. The ledger is immutable, so each decision the agent or a human makes leaves a permanent, reviewable record.
Which AI tools can connect to Pubroo?
- Any MCP client, including Claude and ChatGPT, can connect to Pubroo because Pubroo is an MCP server. Connecting does not grant access on its own: the agent must be granted scopes by a human on a consent screen before any community tool becomes available to it.
Do I have to use the AI moderator?
- No. A Pubroo community works as an ordinary hosted forum with categories, tags, threaded topics, search, roles, and invitations. The agent is additive, and a community with no grant issued has no agent access at all.
Put your brand behind an AI you can audit.
Free tier. Your domain. No credit card.
Create your community