What is a project memory graph?
A project memory graph is a connected record of a team's notes, decisions, tasks, and the relationships between them, structured so an AI can read it and propose work the team should do next. Unlike a flat task list, which just stores items, or a generic AI summarizer, which condenses text into prose, a memory graph links information to its source and uses those connections to surface action items and decisions. In Verkion, the Memory Graph reads your workspace's own notes and proposes suggestions that each cite the source note they came from, so a human can review and approve before anything is created.
What a project memory graph actually is
A project memory graph is a structured, connected representation of the knowledge inside a project: the notes a team writes, the decisions it makes, the tasks it tracks, and the links between those things. Rather than treating each note or task as an isolated entry, the graph captures how they relate, so the meaning of the project lives in the connections, not just the individual items.
The 'graph' part matters. Because information is connected to its origin and to related items, the record can be read by software, including AI, to reason about what a team has discussed and what it should do next. The 'memory' part matters too: as more notes and decisions accumulate, the graph holds more of the project's history in one connected place.
The practical payoff is that work proposals can be grounded in what the team has actually written, rather than invented from thin air or pulled from a generic model's assumptions.
How it differs from a flat task list and a generic AI summarizer
A flat task list is a collection of to-do items with no memory of why they exist. It tells you what to do but not what decision led there, which note raised it, or how items connect. Nothing reads the list back to you and proposes the next item; you add and check off entries by hand.
A generic AI summarizer goes the other direction: it compresses documents into prose. That can be useful, but a summary is disposable output with no durable structure, no links back to specific sources, and no way to turn a suggestion into a tracked, approvable piece of work. You also usually can't tell which note a given sentence came from.
A project memory graph sits between and beyond both. It keeps the structure of a task system, the source-awareness a summary lacks, and the ability to propose connected, traceable work, while leaving the decision to act with a person.
How Verkion's Memory Graph works
Verkion's Memory Graph reads your team's own notes and documents and proposes action items and decisions based on what it finds. Each suggestion is cited to the source note it came from at the document level, so you can open the originating note and see the basis for the proposal rather than trusting an opaque answer.
Verkion is built around a human-in-the-loop model: the AI proposes, and a person approves. Nothing is created in your project until someone approves it, which keeps the team in control of what actually enters the initiative-to-workstream-to-task hierarchy. Suggestions also get sharper as the workspace's notes grow, because there is more connected context to draw on.
AI analysis is included on all paid plans, subject to a per-user fair-use rate limit. Your content is processed to generate these suggestions but is not used to train AI models, and each workspace is isolated from others via Postgres row-level security.
Frequently asked questions
Is a project memory graph the same as a knowledge graph?
They share the same underlying idea of connected, structured information, but a project memory graph is scoped to a single team's project work, its notes, decisions, and tasks, rather than general world knowledge. The goal is not to model everything, but to connect what a team has written so an AI can propose relevant next steps. In Verkion, that scope is one workspace, isolated from others.
Does the AI create tasks automatically from the memory graph?
No. In Verkion, the AI proposes action items and decisions, but nothing is created until a person approves it. This human-in-the-loop approach keeps the team in control, so the graph surfaces suggestions rather than silently adding work. You review each proposal and decide whether it becomes a real task.
How does the memory graph cite its suggestions?
Each suggestion in Verkion's Memory Graph is cited to the source note it was drawn from, at the document level. That means you can trace a proposed action item or decision back to the specific note that prompted it, rather than getting an answer with no provenance. It is a document-level citation, not a verbatim line-by-line quote.
Do I need a lot of notes for a memory graph to be useful?
It helps. A project memory graph draws on the notes and documents a team already keeps, so the more context the workspace contains, the sharper the suggestions become. In Verkion, suggestions improve as the workspace's notes grow. You can start small and let the graph become more useful as your team writes more.