Verification guide

Make sure everything works.

Run these tests in any connected AI tool — Claude, ChatGPT, Codex, Cursor, Lovable, or Gemini. Each one checks a different part of Engrams. Takes about 5 minutes.

0 of 5 passed

Prerequisite: Connect Engrams to at least one AI tool first. See the setup page for instructions.

Basic memory

01 Remember and recall pending

Checks that Engrams can save a memory and retrieve it in a new conversation.

a.Start a new conversation in your AI tool
b.Say: "My favorite programming language is Rust"
c.Wait for the AI to respond (it should call remember automatically)
d.Start a completely new conversation
e.Ask: "What's my favorite programming language?"

Expected result

The AI answers "Rust" — it remembered across conversations via Engrams.

02 Profile memory pending

Checks that permanent facts about you are stored and recalled.

a.In a new conversation, say: "My name is [your name] and I work as a [your role]"
b.Start a new conversation
c.Ask: "What do you know about me?"

Expected result

The AI mentions your name and role. Profile memories persist permanently and load at every session start.

Cross-tool memory

03 Memory works across AI tools pending

Checks that memories saved in one AI tool are available in another.

a.In one AI tool (e.g. Claude), say: "Our API uses Bearer token auth with JWT"
b.Open a different AI tool (e.g. ChatGPT, Codex, or Cursor)
c.Ask: "What authentication does our API use?"

Expected result

The second tool knows about Bearer token / JWT auth — the memory was saved in one tool but is accessible everywhere. One brain, any interface.

Getting started

04 Bring an existing project into Engrams pending

Engrams picks up your existing project context automatically.

a.Open a project where you've done work before (e.g. a Claude Project)
b.Make sure Engrams is connected
c.Start a new conversation
d.The AI will offer to save your existing context to Engrams

Expected result

Your project context is saved to Engrams. Next session — in any AI tool — it all comes back automatically.

05 Invite a collaborator pending

Share a project with someone else. Both of you get the same memory.

a.Tell your AI: "Share this project with name@email.com"
b.They receive an email with an invite link
c.Once accepted, the project appears in their AI automatically

Expected result

Both of you read and write to the same memory. Decisions, tasks, and context are shared. Each memory shows who created it.

Memory types

Engrams uses seven memory types, each designed for a different kind of information. Your AI chooses the right type automatically — you don’t need to specify it.

profile LOADS EVERY SESSION

Permanent facts about you — name, role, preferences, expertise. Saved once, recalled forever. Updated when facts change.

context LOADS EVERY SESSION

Active project state — what’s deployed, what changed, current status. The AI’s working memory for your project.

decision LOADS EVERY SESSION

Choices you’ve made with rationale. Why you picked one approach over another. Prevents the AI from re-opening settled debates.

learning SURFACES WHEN RELEVANT

Insights worth keeping — what worked, what didn’t, patterns discovered. Retrieved via semantic search when the topic comes up.

episode LOADS EVERY SESSION

Session summaries — what was done, what’s next, open questions. Lets your AI pick up exactly where the last session left off.

instruction LOADS FIRST, HIGHEST PRIORITY

Rules and protocols for how your AI should behave in this project. Loaded before everything else. Use for format rules, workflows, or constraints.

task LOADS EVERY SESSION

Structured work items with status, priority, and owner. Create, update, and track tasks across sessions. Your AI knows what’s in progress and what’s done.

How memories work — active vs archived

Engrams versions your memories. When a memory is updated — a task moves from in_progress to done, a decision is revised, a profile fact changes — Engrams writes a new row and marks the previous one as archived. The archived row is not deleted. It stays in the database with a pointer to the version that replaced it.

Active

The current version of each memory. This is what your AI sees when it calls recall or load_project. The dashboard "Active memories" count is this number.

Archived

Older versions that have been replaced. Never returned by recall or load_project. Preserved so you can audit how a decision or task evolved over time.

Why keep archived rows at all? Because the history of a decision is often as useful as the decision itself. If a task was set to done but later reopened, or if a profile fact flipped, the archived row lets you reconstruct what the AI believed at any point.

Does archived count against my plan? No. Plan limits apply to active memories only — updating a task or revising a decision never eats quota.

All tests passed.

Engrams is working correctly. From here, everything is automatic — just use your AI tools as you normally would. Memories build up over time.

Questions? Write to hello@engrams.app