The problem
Everyone on your team re-explains the same context to their AI. One person has the runbook. Another has the decision log. A third has the onboarding notes. None of it is connected. None of it carries across conversations.
The pain scales with headcount. A new hire opens Claude and asks, "How do we handle deploys?" Claude does not know. A founder asks ChatGPT, "What did we decide about pricing last quarter?" ChatGPT does not know. Every conversation starts from zero.
Hjarni gives your team a shared knowledge base. Claude and ChatGPT read from it.
What a team folder looks like
A typical setup is one shared folder per kind of context, each with its own AI instructions:
For example, on Runbooks: "When asked about deploys or incidents, search this folder first and follow the steps exactly." The team rule sits above it, so every team folder inherits a shared baseline:
How teams use Hjarni
Every team member's AI works from the same context. No copy-pasting. No repeating what you told someone else's AI last week. The same architecture note that informs Claude when an engineer asks about the database is the one that informs ChatGPT when a founder drafts the investor update.
AI onboarding for new hires
The hardest part of joining a team is the unwritten context. The reason this service exists. The reason that other one was sunset. The thing the founder always says about pricing. The reason you use Postgres.
In most teams that context lives in three Slack channels, one Notion page, and one person's head. With Hjarni, you write it down once and the new hire's AI reads it on day one.
A new engineer connects Claude to Hjarni and asks, "Walk me through how the billing service works." Claude reads the Architecture folder and the Decisions log, then walks them through it. The senior engineer was not interrupted.
Runbooks the AI can read
A runbook in a wiki is read by humans, slowly, at three in the morning. A runbook in Hjarni is read by Claude or ChatGPT, instantly, when someone asks.
Put your deploy steps, rollback procedures, and incident playbooks into a Runbooks folder. Add a folder-level AI instruction. Your team's AI becomes the first responder.
This is not magic. It is documentation that AI can read. The work is still writing the runbook. What changes is who reads it.
Decision logs as institutional memory
The decisions your team makes are the most valuable knowledge you produce, and the easiest to lose. Why you picked Postgres. Why you moved off Redis. Why pricing went from per-seat to flat. What you tried and rejected.
Write them down once. Put them in a Decisions folder. Tag them by topic. From then on, when anyone on the team asks Claude or ChatGPT "why do we do X this way", the answer comes from your decision log, not from guesswork.
This is the institutional memory most companies lose every time someone leaves.
Shared customer interview memory
If your team talks to customers, the notes pile up fast. Thirty interviews in, no one remembers what Sarah said about pricing or whether two customers mentioned the same workflow gap.
Put interviews in a folder. Tag by theme. Add a folder-level instruction: "When asked about customer pain points, search this folder first and quote interviewees." Now any team member can ask Claude or ChatGPT "what did our last ten customers say about onboarding" and get a real answer with real quotes.
What makes it different
- AI-native, not AI-added. The MCP server is built in, not bolted on. Every folder can have its own AI instructions. Claude and ChatGPT follow them.
- Personal and shared together. Personal notes stay personal. The team knowledge base is shared.
- No bundled AI. Team members bring their own ChatGPT or Claude. No per-message costs. No token limits per seat. No surprises on the bill.
- Simple on purpose. Notes, folders, and a connection to your AI. No databases, kanban boards, or page builders. The constraint is the feature.
When someone new joins
They connect Hjarni to their AI. Claude already knows your conventions and decisions. The context that used to live in someone's head is now written down. Their AI can read it on day one.