How Mimir learns about your company, users, and goals — and why it matters.
For an overview of how Mimir builds understanding over time, see how Mimir learns. This page covers the details of the knowledge system and how to manage it.
Mimir builds a knowledge base about your business organized into seven categories: company profile, competitive landscape, user segments, goals & constraints, product state, terminology, and metrics. This context makes recommendations specific to your situation rather than generic product advice.
The more Mimir knows about your business, the sharper its output becomes. Instead of “improve onboarding,” you get “add a guided setup flow for mid-market SaaS teams who churn within the first week.” Context is what bridges the gap between a general-purpose AI and one that thinks like a member of your team.
Mimir learns about your business automatically as you use it. There is no special setup step — context accumulates naturally from two sources.
Automatic extraction from sources
When you upload an interview transcript that mentions “our target users are mid-market SaaS teams,” Mimir captures that as a user segment. Every source you add is scanned for business details — company facts, competitive signals, metrics, goals — and anything relevant is saved automatically.
Chat conversations
Business details you share in conversation get extracted automatically. Saying “we're focused on reducing churn this quarter” becomes a goal. You don't need to do anything special — just talk naturally and Mimir picks up the signals.
Every piece of business knowledge Mimir captures is classified into one of seven categories. Together, they form a complete picture of your business context.
Company profile
What your company does, your industry, stage, and team size. The foundation that shapes how Mimir frames recommendations.
Competitive landscape
Key competitors, your differentiation, and market positioning. Helps Mimir avoid recommending features that would put you in a weak competitive position.
User segments
Who uses your product, their personas, and primary use cases. This is how Mimir knows which users a recommendation would affect most.
Goals & constraints
Business goals, technical constraints, and timeline pressures. Ensures recommendations are feasible, not just desirable.
Product state
Current features, tech stack, and known issues. Mimir uses this to avoid suggesting things you've already built or flagging problems you're already aware of.
Terminology
Domain-specific terms and jargon your team uses. Helps Mimir speak your language and recognize important concepts in your sources.
Metrics
Key metrics you track, benchmarks, and targets. The more Mimir knows about what you measure, the more relevant its recommendations become.
Each knowledge entry has a confidence score based on how many independent sources support it. If three different interview transcripts mention that your users struggle with onboarding, that user segment insight gets a high confidence score. A single offhand mention in one source gets a lower score.
This helps you see what Mimir is most and least sure about. High confidence means multiple sources agree. Low confidence means a single mention — it might still be accurate, but Mimir is transparent about the evidence level behind each entry.
With context about your business, recommendations become specific. Instead of “improve authentication,” you get “add SSO for enterprise customers.” Instead of “reduce churn,” you get “add a setup checklist for mid-market teams in their first week.”
The AI knows your user segments, constraints, and goals, so it suggests things that actually fit. A recommendation for a 5-person startup looks very different from one for a 200-person enterprise team — context is what makes that difference.
Head to Settings to review all knowledge entries organized by category. Each entry shows its content, confidence score, and which source or chat message it was extracted from. This is where you can see exactly what Mimir knows about your business.
Delete anything that's wrong or outdated — old competitor information, metrics you no longer track, or terminology that has changed. Entries update automatically as you add more sources and chat messages, but cleaning up stale entries keeps the context sharp.
Share context naturally in early conversations
Even five minutes of context-sharing dramatically improves every analysis Mimir runs. When you first create a project, tell Mimir about your company, your users, and what you're trying to achieve. It's the single highest-leverage thing you can do for output quality.
Mention changes when your strategy shifts
If your company changes direction — new target market, new primary metric, new competitive positioning — bring it up in chat so Mimir's knowledge base reflects your current reality.
Richer context = dramatically better recommendations
There is a direct relationship between how much Mimir knows about your business and how useful its recommendations are. Generic input produces generic output.
Delete outdated entries
Old competitor info, deprecated terminology, or metrics you no longer track can mislead the AI. Keep your knowledge base clean by removing entries that no longer apply.