Memory & Knowledge

Core Concept 8 min read Updated May 2026

What is GrooveOS memory?

Every piece of knowledge your team produces — a conversation, a web clip, a Drive document, an agent output — becomes a memory item in GrooveOS. Memory items are stored permanently, tagged automatically, and searchable by any team member or AI tool.

You don't have to think about where knowledge lives. Everything your team produces lands in one place, structured, scoped to your team, and available instantly to every AI conversation you start.

The Tagging Contract

Every memory item in GrooveOS carries 7 automatic tags. You don't set most of these manually — GrooveOS infers them from context.

Tag	What it means	Example
`team_scope`	Which team owns this item	`excalibur`
`project_scope`	Which project it belongs to	`fundraising`
`visibility`	Who can see it: `team` / `project` / `private`	`team`
`confidence`	How confident is the source (0–1)	`0.9`
`truth_level`	How verified is this fact	`WORKING`
`source`	Where this came from	`librechat:conv_abc`
`validation_status`	Pending, approved, or rejected	`pending`

Overriding Tags

You can override any of these when sending via the API or Chrome extension. The defaults are sensible for most use cases.

The 5 Truth Levels

Facts in GrooveOS aren't binary — they exist on a spectrum of certainty. Every memory item carries a truth level that tells you and your AI tools how much to trust it.

EPHEMERAL

Raw capture

A quick note, a web clip, a first draft. Unverified. May expire if never promoted.

WORKING

Actively used knowledge

In-progress facts, current best guesses. The default level for most inputs — your AI uses these but treats them as provisional.

VALIDATED

Peer-reviewed

A team member has explicitly approved this fact. Safe to reference in decisions and share with clients.

CANONICAL

Team truth

The definitive answer your team has agreed on. Automatically injected in every AI conversation. This is the level that matters most.

PUBLIC

Published

Approved for sharing outside the team. Requires explicit promotion from CANONICAL by an admin.

CANONICAL facts are special

Every time a team member starts a chat with Claude or GPT-4, GrooveOS automatically includes your team's CANONICAL facts in the system prompt. Your AI always knows what your team knows — without you having to paste it in every time.

Promoting Facts

Facts move up the truth ladder through a human-approval workflow:

Any member can request a promotion: in LibreChat, click the memory item → "Propose as VALIDATED".
An admin reviews the promotion request in Open WebUI → Approve or Reject.
On approval, the fact's truth level updates and it becomes available at the new level across all AI conversations instantly.

You can also trigger promotions via the API:

# Request promotion to VALIDATED
curl -X POST https://api.grooveos.app/v1/promotions \
  -H "Authorization: Bearer $YOUR_TOKEN" \
  -H "X-Team-Scope: your-team" \
  -d '{
    "item_id": "mem_abc123",
    "target_level": "VALIDATED",
    "justification": "Confirmed with client on 2026-05-01"
  }'

Searching Memory

Search your team's memory from LibreChat, Open WebUI, or the API:

From LibreChat

Type /search [query] in any conversation, or just ask the AI directly — it searches your team memory automatically before answering.

Via the API

curl "https://api.grooveos.app/v1/memory/search?q=fundraising+target&truth_level_min=WORKING" \
  -H "Authorization: Bearer $YOUR_TOKEN" \
  -H "X-Team-Scope: your-team"

Results are always scoped to your team. Other teams' data never appears in your results — isolation is enforced at the infrastructure level.

Memory Item Limits

Plan limits

Free plan: 1,000 memory items. Team plan: unlimited. When you reach the free limit, new items are rejected until you upgrade or delete old items. You will receive an email notification at 80% and 100% capacity.

Need help managing your memory items? See Plans & Billing for upgrade options, or API Reference for bulk deletion endpoints.