Workspace Memory

Workspace memory is the context agents retain between conversations — notes about your environment, your team’s preferences, and what happened recently. Agents open each new conversation already carrying it, so you don’t re-explain your setup every time.

​

Why workspace memory

• Less repeated context. An agent that remembers your naming conventions and exempted resources doesn’t ask about them twice.
• Continuity across days. Each night CloudThinker rolls the day’s conversations into a workspace recap, and the next day’s conversations start aware of it.
• A shared team brain. Memory belongs to the workspace — the shared space your team works in (glossary) — so what one member teaches an agent benefits everyone.
• Suggested next steps. Unfinished action items from recent conversations come back as one-click prompt suggestions in the composer.
• Nothing is forever. A memory that goes unused for 90 days is archived and stops informing conversations.

​

What agents remember

Agents write memory themselves during normal work — CloudThinker does not record raw transcripts into memory, and there is nothing to configure. Specialized memory builds on the same foundation: Incident Memory keeps lessons from past investigations, and Learnings keeps reviewer feedback for Review.

​

Prompt suggestions in the composer

When recent conversations left clear next steps, the empty composer shows up to three suggested prompts, each tagged with the conversation it came from. Suggestions draw on roughly the last week of workspace activity, and typing a few characters filters them to match what you’re writing.

​

Ask about past conversations

Agents can find and read earlier workspace conversations, so prior work becomes evidence in the current chat. Find a conversation. Ask an agent which earlier chats mentioned a resource, or what the team worked on last Tuesday. Search covers conversation titles, your messages, and visible agent replies in the current workspace — not tool output, agent reasoning, or attachments. Results return short, redacted snippets as leads. Read a specific conversation. Paste a prior CloudThinker chat link or conversation ID and ask the agent to continue from it. The agent reads that conversation’s transcript in bounded excerpts and treats it as evidence — it never replays old instructions as commands. Every conversation type works, including assessments, code reviews, pipeline analyses, and Learnings. Both stay inside the current workspace: agents cannot search or read conversations from another workspace, and deleted conversations are excluded.

​

Correct or remove a memory

Memory is agent-maintained, so you fix it the same way it was written — in chat:

1. Tell any agent what’s wrong: “Forget that we exempt the staging cluster” or “Update your note — reports go to #finops now, not email.”
2. The agent updates or removes the note immediately, and the correction applies to every later conversation in the workspace.

Memories no agent has used for 90 days are archived automatically, so stale context ages out even when nobody corrects it.

​

FAQ

​

Related