Instructional infographic explaining Persistent Project Intelligence: how session receipts are turned into a knowledge graph and vector memory, retrieved on demand, and used to rehydrate context for new AI collaborators.

Persistent project intelligence

How ACCPT turns session history into memory the next collaborator can use
Documentation records information. Project intelligence organises it into knowledge.
1
Capture the session
Every completed session ends in a structured receipt, not a raw transcript.
2
Extract the knowledge
The receipt is parsed: entities, decisions, risks and open questions are pulled out.
3
Store it two ways
Neither representation works alone — together they cover facts and meaning.
Knowledge graph
Explicit relationships — what depends on what, and why.
Vector memory
Semantic similarity — related discussions, even in different words.
4
Retrieve on demand
A context retrieval layer assembles what's relevant to the task at hand.
5
Rehydrate the next contributor
A new collaborator starts from an assembled context package, not from zero.
Requirement · ACCPT-PPI-001
Every session receipt must record
Objectives completed
Architectural decisions
Assumptions
Risks
Outstanding work
Files modified
References to supporting artefacts
Without PPI
Manually reads the README, architecture docs, build logs and past receipts.
With PPI
Retrieves relevant decisions, constraints and open issues automatically.
Both paths still end in the same Declaration of Understanding — only how it's prepared changes.