The shared memory behind every Hamster answer — your Briefs, conversations, decisions, documents, code, and the people on your team, connected so the AI can reason over them as one workspace.
Overview
Hamster keeps a live picture of your workspace called the Context Graph. It is the layer the AI assistant reads from when it answers a question, drafts a Brief, or generates a plan. Instead of a flat pile of documents, the graph captures how your work is connected: which Brief belongs to which Initiative, which Task lives under which Brief, which conversation produced which decision, and which document was attached to which thread.
That structure is what lets the AI walk relationships when it answers you. Ask "what Briefs are linked to the onboarding Initiative?" and Hamster does not have to keyword-match against document chunks — it can read the relationships directly and ground the answer in the actual shape of your workspace.
This page is the home for how the graph works and how it shows up across Hamster. It will grow as the graph gains more surfaces and capabilities.
What Hamster Gathers
The graph captures the entities your team works with and the relationships between them:
Code repositories — Connected GitHub repositories, with code symbols (functions, classes, etc.) extracted from 27 programming languages and indexed for search. Each repository can track a different branch, and the AI understands which branch each code snippet came from.
Goals — Organizational goals and how they relate to Initiatives, Briefs, and team members.
Ownership and participation — Who owns a Brief, who owns a goal or Initiative, who participated in a thread, who is responsible for a Task, and who voted on alignment.
Conversations and threads — The chats that produced Briefs, the questions that led to a decision, the threads attached to specific Briefs and documents, and the participants in each.
Documents — Strategy decks, PRDs, customer research, transcripts, and any file your team uploads or links into a brief.
Activity and reactions — Alignment votes, status changes, key actions that move work forward, emoji reactions on messages, and who participated in decisions.
When you upload a file, attach a document, or create a Brief, Hamster does not just store the file. It also indexes the content so passages — not just whole files — can surface when they matter, and it records how that file relates to the rest of your workspace.
How It Stays Current
The graph is a live system, not a one-time index:
Realtime updates — When a teammate edits a Brief, attaches a document, or moves a Task, the graph updates within seconds. The AI assistant sees the new state on its next reply.
Backfill on connect — When you connect a tool or upload a batch of context, Hamster works through the content in the background and adds it to the graph as it processes. You can keep using Hamster while it catches up.
File content extraction — Uploaded PDFs, Word docs, and similar files are parsed for their text content, not just their filenames, so the AI can search inside them as easily as inside a brief.
Code indexing — When you connect a GitHub repository, Hamster automatically indexes code symbols from 27 languages and makes them searchable. Code changes are picked up on the next sync; you can manually re-sync at any time.
You do not need to do anything to maintain the graph. There is no manual indexing step, no "rebuild" button, no separate tool to manage. When you connect a GitHub repository, Hamster automatically indexes code from 27 languages. As your workspace evolves, the graph evolves with it.
How Retrieval Feeds the AI's Answers
When you ask the AI a question — in a Brief chat, a thread, or a Plan generation — Hamster does not just stuff a generic chunk of text into the prompt. Retrieval works in three complementary ways:
Direct entity reads. If you mention a Brief, an Initiative, or a document by name or link, the AI reads it directly. No fuzzy matching needed when the answer is in a known place.
Relationship-first lookup. When the question spans connected work — "what's the latest on the Briefs in this Initiative?" — the AI walks the relationships between entities first. This catches Briefs that should be relevant even when they do not share keywords with your question.
Semantic search across passages. When the question is open-ended, the AI searches across passages (the chunks of text inside Briefs, documents, and threads) and returns the most relevant ones along with where they came from. Each match carries enough context for the AI to cite it back to you.
For multi-step questions, Hamster's planner can chain these together — read a Brief, expand to its linked Initiative, search across the documents attached to both — and stay within a sensible budget so answers stay fast.
Where You Will See It
Chat — Replies in any conversation can reference past threads, linked Briefs and Tasks, attached documents, code snippets, and the relationships between them. See AI Responses.
Documents and context — Uploaded files and linked documents are indexed into the graph. Document passages can map to different parts of the graph so the right section, not just the right file, can surface. See Context Documents.
Plans — Plan generation pulls in the related Initiatives, prior decisions, code context, and documents that should inform the Plan, not just the Brief in isolation. See Generating Plans.
Capabilities
Relationship-first retrieval — The assistant prioritizes context that is linked to your current work, not just textually similar.
Passage-level grounding — When the AI cites a document, it cites the specific passage, not the whole file.
Ownership-aware context — The graph tracks who owns each Brief, Initiative, and goal. Ownership relationships influence which linked context surfaces first.
Code-aware search — The AI understands code from 27 languages and can correlate code snippets with the Briefs or issues they address.
Team- and time-aware context — Recent activity, ownership, and who is involved can influence which linked context surfaces first.
Traceability — You can reason from outcomes back through execution, Briefs, and the conversations and decisions that led there.
Account-isolated — The graph is scoped to your account. Queries always run within your team's data — no risk of context leaking across workspaces.
Tips
The richer the inputs, the better the retrieval. Upload customer research, attach strategy docs, link Briefs into Initiatives, and connect GitHub — every relationship gives the AI another path to find the right context.
If the AI's answer feels off, check whether the right context is actually linked. Often a missing Brief→Initiative link or a forgotten document upload explains why a relevant passage did not surface.
Hamster keeps file content searchable, so naming a file well matters less than making sure its content is in the file. The AI reads the content, not just the title.
Newly uploaded documents take a few seconds to a few minutes to index, depending on size. The status indicator on the Brief tells you when processing is complete.
Code indexing happens automatically after GitHub connection. You can manually trigger re-sync from the Connections page.