Memories and search

Studio has two related memory layers:

• Operational memories are facts you or Copilot save for later recall.
• Organization memory records durable organization facts and work products such as generated apps, artifacts, repositories, deployments, or services.

Unified search finds these alongside hosts, files, conversations, procedures, dashboards, connectors, and commands.

​

Operational memories

A useful memory changes how the next person approaches the work. It is specific, scoped, and understandable without the original transcript. Save a memory from chat by asking Copilot, or create and edit one from the memory surface when available through search and AI settings. Update a fact when reality changes instead of adding a contradictory duplicate.

​

Categories and tags

Operational memories can use categories such as general, topology, procedure, credential hint, incident, and configuration. Tags add cross-cutting retrieval for a customer, site, ticket, protocol, migration, or time period. A credential memory may say which named credential to use. It must never contain the credential value.

​

Organization memory

Organization memory stores a durable record with a stable subject identity. Copilot can create or update records for:

• Organization facts.
• Applications and dashboard apps.
• Artifacts and documents.
• Repositories and workspaces.
• Deployments and services.

A work-product record can include a title, summary, last-known state, locator, source host, source session, workspace reference, revision, framework, process, port, and searchable tags. Studio strips credential-like metadata before persistence, but you must still avoid placing secrets in the title, summary, locator, tags, or metadata.

​

Stable subject IDs

When a work product changes, update the existing record using the same stable subject identity. For a machine-built app, the absolute workspace path is a strong identity. Creating a new record for every edit produces duplicates and makes later retrieval ambiguous.

​

What not to store

Do not store:

• Passwords, tokens, private keys, recovery codes, or session cookies.
• Short-lived values that should come from a live system.
• Unverified AI conclusions stated as fact.
• Large transcripts that belong in a conversation, replay, or artifact.
• Personal or performance information unrelated to operations.

Use Key Chain for secrets, the active tool for live state, and a dated artifact for a complete record.

​

Unified search

Press Mod+K to search Studio. Results can include content and registered commands. Search uses names, text, metadata, and semantic relevance, so related terms can surface an item even when the exact phrase differs. Search before creating a new object. Reusing the existing host, procedure, dashboard, memory, or app preserves ownership, history, and stable identity.

​

Search patterns

​

Retrieval in Copilot

Ask Copilot to search before answering when prior context matters:

• “Find the latest verified memory for this site’s VPN before troubleshooting.”
• “Locate the existing dashboard app for WAN capacity and verify its current revision.”
• “Search for the procedure and incident report from the last carrier outage.”
• “Update the existing organization record after verifying the deployment.”

When a memory directly affects a recommendation, Copilot should identify the remembered fact and distinguish it from live evidence.

​

Memory hygiene

• Include scope: organization, customer, site, host, interface, or service.
• Record when the fact was verified and by what evidence when that matters.
• Use one stable record per work product.
• Update or archive stale records.
• Separate historical findings from current state.
• Review shared memories before teammates depend on them.

​

Related