Shared Agent Memory

Keep the knowledge from your AI work available after the chat ends. MindWiki gives multiple agents access to the same saved briefs, decisions, research, and handoffs, so a new session can retrieve the context it needs instead of asking you to rebuild it.

The most useful memory is visible, editable, and specific. You can open it yourself, correct it, link it to evidence, and ask a different agent to continue from it.

Use a simple memory loop

1. Retrieve: read the current brief and relevant decisions before starting.
2. Work: investigate, write, design, or build using that context.
3. Save: preserve the result, reasoning, evidence, and remaining work.
4. Continue: ask the next agent to retrieve the saved handoff.

A connection makes this possible. The habit of saving and retrieving makes it useful. Connecting an agent does not automatically import its past chats or guarantee it will read the right page without a task instruction.

Save what will change the next session

Save a concise working record rather than an unexplained transcript. Preserve exact details when they matter, such as a file path, requirement, date, or verification result.

Start a session with context

Prepare MindWiki context for the Website Launch project. Read its current brief, decisions, and latest handoff. Summarize the goal, constraints, and next action, citing the pages used, before starting work.

Use a project hub for a small set of linked pages. Create a Context Pack when the same sources repeatedly support the work. The context router can choose a pack and supplement it with retrieval when the task needs more.

End with a useful handoff

Save a concise handoff to MindWiki. Include what we completed, what was verified, decisions and reasons, source pages, unresolved questions, and the next concrete action. Read any existing handoff before updating it and tell me the saved path.

A reusable page structure is:

# Website Launch — Handoff

## Current goal
What we are trying to accomplish.

## Completed and verified
What exists now, with the evidence or check performed.

## Decisions
What we chose and why.

## Open questions
What remains uncertain.

## Next action
The first useful step for the next session.

## Sources
[[Website Launch Brief]]
[[Homepage Messaging Decision]]

Give different agents clear roles

A research agent can gather and summarize sources. A planning agent can turn those findings into requirements. A coding agent can retrieve the requirements and save implementation results. A writing agent can use the same evidence for an announcement.

Each client connects independently to your existing account. This shares knowledge, not live conversational state. Tell the next agent which project or handoff to retrieve and what task it should perform.

Keep shared work current

Ask agents to read before editing an existing page. Updates use the current content hash so a stale agent cannot silently overwrite a newer revision. If two agents work on the same page, a conflict means they need to reread and reconcile the current content.

For parallel tasks, separate output pages are often clearer: one research note, one draft, one implementation handoff, all linked from a project hub. Bring the results together after reviewing them.

Preserve human control

Open important pages yourself and correct summaries that drift from your intent. Keep source notes linked to conclusions. Use proposals for supported organizational changes and review recorded activity in the Mac Agents workspace.

Folder names and instructions help agents follow your conventions; enforced scopes and vault permissions control what they may do. Revoke an individual grant when a client no longer needs access.

Next: Project Workflows, Context Packs, and Mission Control.