Knowledge Hub
Drafts that sound like your team wrote them.
The Knowledge Hub holds what your agents must get right about your business: who you sell to, what you are working toward, how you sound, what you never say, and which things share your name but are not you. Every sort, draft, and check reads from it.
Amara Osei
linkedin · 5h
"We're evaluating pipeline tooling for a platform team of 40. We care about audit logs and SSO."
Draft grounded in your docs and Knowledge Hub
Both are built in. Audit logs cover every workspace action, and SSO works with SAML and OIDC. Happy to walk your team through a setup that fits 40 engineers.
A person approves before anything is sent.
What it holds
Six kinds of knowledge, written plainly.
Who you sell to
Relevance is scored against your profiles, not a generic developer.
Platform teams, 20 to 200 engineers, self-hosting or air-gapped.
Objectives
Posts that bear on a goal rank higher.
Win the enterprise segment this quarter.
Traits
How the product must come across in every draft.
Plain and direct. No hype words.
Guardrails
A draft that breaks one is stopped with the reason attached.
Never promise roadmap dates.
Capabilities
What the product actually does, so drafts describe it as it is.
SSO with SAML and OIDC.
Name collisions
Things named like you that are not you. Their posts are screened out.
ACME token (crypto) and Acme (indie band)
How it works
Read before every decision. Checked after every draft.
1. Screen
Ears agent: judges each post against your confirmed profiles and name collisions before it reaches the queue.
"ACME token just pumped 40% today"
MutedName collision: ACME token2. Draft
Ears agent: reads the relevant part of the brain alongside your docs, writes the draft, and cites both.
"SSO works with SAML and OIDC, and audit logs cover every workspace action."
docs/sso.mdCapability: SAML and OIDC SSO3. Check
Ears agent: tests every draft against your guardrails. You: read it, edit it, approve it, or skip it.
"This ships next month"
Blocked: Never promise roadmap dates
4. Learn
Ears agent: your edits and skips become proposed lessons. You: confirm or reject each one before Ears uses it.
"Check out our blazing fast feature" "Set up your first source in two steps"
Lesson candidate: Lead with the setup step, not the feature name
ApproveReject
Under your control
Versioned, inspectable, removable.
Nothing enters the active brain without a person confirming it. Extracted items start as candidates you approve or reject. Every change keeps the version before it, so you can see what the agents knew at any point. Delete an item and its whole history goes with it, and the agents stop using it immediately.
Guardrail: Never promise roadmap dates
v2 · Edited by Priya · Sep 18
v1 · Confirmed by Dev · Sep 2
Candidate · Extracted from docs/roadmap-policy.md · Sep 2
FAQ
Common questions
How does context get into the Knowledge Hub?
Three ways: paste text such as a doc or positioning note, write a rule directly, or point at a page such as your docs or site. EarShots reads them into candidate items that a person confirms before the agents use them.
Can the agents change the Knowledge Hub on their own?
No. Agents propose candidates and lessons from outcomes. A person approves or rejects them. Confirmed rules can only be changed by a person, and every change is versioned.
What happens when I delete something?
The item, all of its historical versions, and every relationship touching it are removed. The agents no longer use it from the next decision onward.
Why do name collisions matter?
Common product names pull in posts about bands, coins, and unrelated tools. The Knowledge Hub stores each namesake with the lines that distinguish it, and those posts are screened out before they reach your queue.
Give your agents a memory.
Add your docs and a few rules. The next draft follows them.