← CLA · CommandKnowledge & LearningSelf-improvement →

CONNECTED WORKSPACES

Your conversations. Shared, traceable knowledge.

Connect a Claude Code workspace to one agent. It retrieves authorized context before each prompt and lets you publish selected findings. Each conversation keeps its own history.

Verify your work identity

Email-only Hub login does not authorize source access or desktop pairing.

Preparing Microsoft sign-in…

Connect Claude Code

Use a separate connection for each agent or workspace. Existing agent permissions apply. Microsoft sign-in is required.

  1. Download the connector and inspect it. Requires Node.js 20 or newer.
  2. In Terminal, run node ~/Downloads/cla-sync.cjs pair.
  3. Paste the code, then use the printed claude --settings … command for that agent. Use the printed resume command to select an existing conversation.

Existing Claude Code windows are unchanged until you start or resume them with their connection settings.

How knowledge travels

  1. Retrieve

    Before a prompt, get a fresh, bounded snapshot of authorized inputs, memory, decisions, proposals and available sources. The hook never sends your prompt or transcript to the Hub.

  2. Contribute

    Explicitly upload a selected source or publish a finding with source hashes. Repeated identical submissions return the existing receipt.

  3. Review

    The same user’s agent sees the proposal as unverified. Source validation and existing human approval gates still apply before it becomes shared policy.

Finding file format
{
  "title": "A finding to verify",
  "finding": "Observation, supporting evidence, recommended action and expected result.",
  "evidence_refs": ["sha256:…64 hexadecimal characters…"]
}

Upload a selected source below or use the connector’s explicit upload command. Reference the receipt’s content_ref in your finding. A matching hash verifies stored bytes, not the truth of a claim.

A context read is not evidence that Claude used it. Pairing, retrieval, model use, verified outcomes and approved learning are different steps.

Your connections

Loading connections…

WorkspaceAgentConnectionLast context returnedLatest errorAction

Rename changes only the workspace label; it keeps the connection and its expiry. Revoked connections are removed from this list and retained in the audit history. Connections expire after seven days. Revocation blocks future reads and contributions; it cannot erase context already delivered to a local conversation.

Selected sources & evidence

Files here are private to you and the selected agent. Indexing makes a source findable; its claims remain unverified. No prompt, transcript or folder is collected automatically.

UTF-8 TXT, Markdown, CSV or JSON; maximum 128 KiB. CSV and JSON are indexed as text, without certifying their figures.

Search covers your selected agent’s indexed sources. Only the terms you submit are sent.

Choose an evidence agent.

SourceStateReceivedEvidenceActions

Withdrawal stops new retrieval and marks linked proposal evidence withdrawn. It cannot remove text already delivered to a conversation. Approved lessons and business outcomes do not increase merely because files were uploaded.