Key naming conventions

Hive stores memories under a key you choose. The server doesn't enforce any structure on that key — any non-empty string works. But keys left to grow organically become chaotic and collision-prone fast, and renaming them later is painful once a client has thousands.

This page describes the recommended convention. It's a guideline, not a rule; adopt as much or as little as suits your agent.

The convention

{domain}:{entity-type}/{entity-id}:{attribute}

• domain — top-level namespace (e.g. project, user, session, team, global)
• entity-type/entity-id — identifies a specific thing (optional, omit for domain-wide memories)
• attribute — what the value contains (e.g. summary, preferences, context)

Lowercase, hyphens within segments, colons as separators, / between an entity type and its id.

Examples

Key	Meaning
project:task/42:summary	Summary of task 42 in the project domain
user:profile/alice:preferences	Alice's user preferences
session:current:context	Context for the current session
team:shared:coding-guidelines	Team-wide coding guidelines
global:env:database-schema	A globally-shared piece of context

Why this works

• Collision-resistant. Two different domains never share a key space.
• Queryable by prefix. Pairs well with list_memories and search_memories when you want everything under project:task/42.
• Readable. An operator browsing the Memory Browser can tell what each memory is for at a glance.
• Extensible. Adding a new domain or attribute never forces a rename of existing keys.

Tips

• Prefer a fixed, small set of domains per agent — don't invent a new one for every memory.
• Keep entity-id stable for the life of the entity (a database id works well; a slug is fine if it never changes).
• Use tags for cross-cutting groupings ("decision", "open-question") rather than stuffing them into the key.
• When in doubt, start simple (domain:attribute) and add structure only when a real collision shows up.

What Hive does not enforce

The server accepts any key. This convention lives in your agent's system prompt or tool-use instructions — it's your choice how strictly to follow it. Hive may introduce opt-in key validation in a future release.