Information is everywhere. Understanding is nowhere.
Here is what your business information looks like right now. Client conversations live in email. Decisions live in your memory, or nowhere. Meeting notes live in a document app, maybe. Action items live in one of three to-do lists. Contact context lives in a CRM, if you use one. Strategic thinking lives in voice notes on your phone.
All of it exists somewhere. None of it is connected.
You know that the decision about the Munich project relates to the conversation with Thomas, which connects to the Q3 timeline, which affects the commitment you made to Markus. That connection exists in your head and in no system. So when someone asks about Munich, you are the one who reconstructs the picture — from five sources, every time. Your information is not disorganised. It is scattered, and the scatter is what costs you the hours.
What an AI knowledge graph is
A knowledge graph is fundamentally different from files, folders, and search. Instead of storing documents, it stores relationships.
A concrete example. You mention “Marcus from the Munich project” in a voice note. In a folder-based system that note is filed somewhere and maybe you find it again. In a knowledge graph it creates connections:
- Marcus (person) → works at → Müller GmbH (organisation)
- Müller GmbH → is the client on → Munich project
- Your voice note → mentions → Marcus, the Munich project, and the decision you described
- Marcus → is linked to → his previous emails, past meetings, and every other note about him
Now “what is the latest on Munich?”, “when did I last talk to Marcus?”, and “what is pending with Müller GmbH?” are the same lookup from different ends. Not search results. Context.
The “AI” part is that the graph is extracted and maintained by models rather than by you: entities recognised, relationships inferred, identities merged, and the whole thing kept current from the stream of your actual work.
What is in PILOT’s knowledge graph
PILOT is a private AI chief of staff, and the knowledge graph is the standing model of your world that everything else runs on. It is built for one principal, and it holds more than nodes and edges:
Real identity resolution. People arrive under different names, addresses, and aliases. A five-rung ladder — email, alias, and embedding-based matching among them — resolves “Thomas”, the Teams invitee, and the signature on a forwarded PDF to one person.
Relationship warmth that cools with time. Every contact carries a warmth that rises with interaction and fades with silence, so the graph can say which relationships are going cold before you notice.
Temporal fact history. Facts have dates. “Who was CFO at Müller GmbH before Carsten?” is answerable, because the CFO → COO switch is stored as a change, not an overwrite — and a role change on a contact surfaces under “Pilot noticed”.
Commitments and decisions as first-class objects. Promises in both directions, with owners and deadlines, and a decisions log with the reasoning attached.
A map. Contacts are geocoded on a world map with a travel radar: you are in Munich on Tuesday, three warm contacts are there, one is going cold.
How the graph builds itself
Nobody maintains it by hand. PILOT ingests Microsoft 365 or Google mail and calendar, Teams chat and meeting transcripts, Zoom recordings as an option, voice notes from your phone or a Plaud recorder, Telegram and WhatsApp messages, photos and PDFs, and business-card photos, which become contacts without the photo being stored.
Every source is read for people, organisations, projects, decisions, dates, and commitments, and for how they relate. A ninety-second voice note after a client dinner — “Markus is worried about the Q3 timeline, his new CTO starts in April, they are evaluating a competitor for the backup system” — becomes a fact on the project, a follow-up to schedule, and a competitive signal on the account, all linked. The graph gets richer every week, and “What Pilot learned this week” in the weekly review shows you the diff.
The questions you can now answer
Once the graph has a few months of your work in it, the questions it answers are the ones a well-informed colleague could:
- “What is our full history with this client?”
- “What did we decide about pricing last quarter, and why?”
- “Who on my side has the most context on the Munich project?”
- “What do I owe people this week, and what am I owed?”
- “When did we last discuss the partnership with Company X?”
- “Which warm contacts are in Berlin next week?”
Ask in chat, by voice, or from Claude or ChatGPT through the MCP connector — the same graph answers everywhere.
Why this is the foundation
The knowledge graph is not a feature. It is what makes every other part of PILOT work. The morning briefing is accurate because it draws on the graph. Email triage is right because the graph knows who matters to you. Meeting prep is a lookup rather than a research task. Decision tracking is useful because decisions link to everything they affect. Without it, an assistant is a chatbot with a good memory for the current conversation and none for your world.
The guide on what an AI chief of staff is explains why memory across channels and across time is the category’s actual definition.
Private by architecture
The graph is yours alone. Each client’s data lives in its own private database schema — no shared tables, not a customer-ID filter — with field-level AES-256 encryption, secrets in Azure Key Vault, EU-hosted on Microsoft Azure. You can export the whole graph as JSON, and deletion is a schema drop, not a flag.
PILOT was built by its founder because every tool he used was an island, and it runs on his own working life every day. The knowledge graph is the bridge.
