MemoryWorks remembers every incident, decision, owner, and dependency your team already learned — and briefs people and AI agents before they change something.
Learns from the tools your team already uses · briefs the agents you already run
Before it acts, it asks MemoryWorks. The answer isn’t a list of search results. It names the constraint the change would break, the outage it would repeat, and whether to stop and get a person.
agent › rotate the token signing keys→ get_orgmemory_briefing(task=rotate the token signing keys
, service=auth-service
)Code, pull requests, threads, and documents become one memory. Every fact points to where it came from — and when something changes, the old truth is kept and marked superseded, not overwritten.
Every briefing ends in a call an agent can act on.
Who owns a service now, who did before, and who to wake up.
What else a change touches: services, shared databases, downstream jobs.
Each briefing’s result is recorded, so the next one knows what worked.
Sources become kinds of memory; memory briefs the agents; outcomes flow back in. Drag the graph, or hover a node to see what it carries.
One time-aware memory, every fact tied to its source.
Anyone can index the same Slack and the same repositories. What only your workspace builds up is the record of which context actually led to the right action.
The agent says what it is about to do and gets back the incidents, decisions, and constraints that apply — each with its source.
Anything consequential waits for a human: inline in the chat, or in Approvals. The verdict is advice, never authorization.
The agent does the work in the editor or tool it already runs in. MemoryWorks never acts on its own.
What happened goes back into memory, so the next briefing knows what worked and what didn't.
An MCP server for assistants, a Python SDK and CLI for everything else, and the same API underneath.
{
"mcpServers": {
"memoryworks": {
"command": "mcp_server/.venv/bin/python",
"args": ["mcp_server/server.py", "--transport", "stdio"],
"env": {
"MEMORYWORKS_API_URL": "https://memoryworks.app",
"MEMORYWORKS_API_KEY": "om_live_..."
}
}
}
}Agents can read, cross-check, and propose. Changes to memory or a connected tool wait for approval.
People and agents only see what the underlying source already lets them see.
OAuth tokens never reach the browser or a model.
Each line links to the memory and the pull request, thread, or document behind it.
Connect one source and ask. Sign in with GitHub and your repositories are ready to pick.