Project Understanding

Status: ✅ Current · Last reviewed: 2026-06-18

What it is

Project Understanding is a living, always-current snapshot of what your project is and what matters in it — its purpose, the key decisions that shaped it, the rules it operates under, what's being worked on now, and where its own knowledge is thin.

It's not a document you maintain. Memeri assembles it from the work you and your agents already do — and you and every connected AI read the same picture.

Four lenses, one model

Understanding has one underlying model with four views (the view parameter):

• current (default) — what's moving now: current focus, recent evolution, coverage.
• technical — how it's built: architecture, operating principles, known constraints, key decisions, entry-point docs.
• strategic — what it is and why: identity, purpose, project story, current focus, key decisions, open questions.
• full — everything.

A coverage section runs through all of them, honestly flagging where understanding is thin rather than inventing detail.

How it's built and kept fresh

Two layers:

• Derived layer (current focus, decisions, constraints, recent evolution, coverage) — assembled fresh on every read from data you already create: initiatives/jobs, work updates, recorded decisions/constraints (Intelligence), the project timeline, documents. Never cached, always current.
• Narrative layer (identity, purpose, architecture, story, principles) — written by an agent or by you in the UI, and cached until updated.

Memeri makes no AI calls of its own to build this: the derived layer is deterministic assembly; the narrative layer is agent-written. It reflects the recorded record of work, not your raw source code.

The agent tool: get_project_understanding

An agent calls get_project_understanding at the start of working on a project to orient fast. It returns a compact, structured map — identity, focus, decisions, constraints, recent evolution, entry docs, coverage — where every item has an id the agent can drill into.

It's a map/index, not the whole archive: get the shape first, then fetch deeper context only where needed. The view parameter picks the slice.

Where you see it

On the Project page → Home lens, as a calm "Project Understanding" card:

• A one-line identity
• Current focus
• Key decisions
• A coverage summary
• A status strip of which narrative sections are written / locked

A badge flags when it's changed since you last reviewed it. "View full understanding →" opens a drawer with the full narrative (editable + lockable section by section) plus read-only derived detail.

How it differs from neighbours

The flow: understand → context → drill → work.

Why it's fundamental

AI work is fragmented across chats, tools, and agents that forget and never share a mental model — so every session starts cold. Project Understanding gives every connected AI the same durable, source-backed picture, so none starts from scratch.

Memeri doesn't just store memory — it maintains a living understanding that every connected AI contributes to and benefits from, and it improves as it's used: a gap one agent finds becomes recorded knowledge the next agent inherits.

Good to know (and what it is NOT)

• Not a recommendation engine. By design it never says "what to do next" or pushes signals. It orients; it doesn't advise.
• It does NOT read your source code. It reflects recorded work — decisions, jobs, updates, docs. If something wasn't written down it won't appear; coverage flags that area as thin instead of inventing it.
• Narrative sections can be empty until someone (an agent or you) authors them.
• "Updated since you reviewed" is per-browser (local) and resets when you open the full view.

Still being built

• Not yet auto-injected into agents at startup — an agent calls the tool explicitly for now (a deliberate gate).
• The "what changed" signal is a badge, not yet a full diff.
• Some narrative sections are first drafts.