Skip to content

Memory Graph

Lithora builds a connected memory of your work. Search in plain language and the agent surfaces the related tasks, PRs, docs, decisions, and people as a navigable graph.

memory-graph0/8 linked
agent
context that compounds as you ship
issuePRdocdecisionperson

A search hub with its related work, animated as a graph — click a node to jump straight to it.

How it works

  • Semantic search: Ask for a topic, not exact keywords — results are ranked by meaning across every surface.
  • Connected context: Each result links to the tasks, PRs, docs, and people around it, so you see the whole thread of work, not one isolated item.
  • Agent recall: The AI Workspace draws on this graph to answer questions with the right context already in hand.

What gets connected

The graph is built from the work you already do in Lithora and the GitHub activity linked to it. Nodes are the things you work on; edges are the relationships between them.

Tasks

Work items with their status, owners, and history.

Pull requests

Linked PRs, reviews, and CI outcomes from GitHub.

Docs & notes

Workspace documents and the context written around work.

Decisions

The calls the team made and why, kept next to the work.

People

Who owns what, and who has touched a given thread.

Searching the graph

Phrase a search the way you'd ask a teammate — by intent, not by remembering the exact title. A few examples:

  • “what did we decide about the billing migration?”
  • “everything related to the auth rewrite”
  • “who has worked on the checkout flow?”

From any result, follow an edge to jump to the connected task, PR, doc, or person — the graph is a map you navigate, not just a list.

The agent uses it too

The same graph is what powers Workspace Q&A. When you ask the agent a question, it pulls the relevant nodes as context and answers with its sources cited, so you can click straight through to the work it drew from.

What it is — and isn't

The Memory Graph surfaces connections across work you already have; it isn't a separate database you edit by hand. Results are ranked by semantic similarity, so treat them as the best-related items to explore, not an exhaustive query result.

Availability

Semantic memory search requires an active Pro or Scale plan, and the check runs on the API — it cannot be worked around from the client. The Memory view also unlocks on a usage milestone: complete 5 tasks and it appears in the navigation, because a graph built on an empty workspace has nothing to show. Find it at Dashboard → Memory, or as the Graph tab of Insights.