User Context

Individual preferences, conversation history, and personal workflow patterns. Each user gets a personalized automation experience layered on top of the shared Business Memory.

What is User Context?

User Context is the per-user slice of memory inside a workspace. Where Business Memory answers “how does this company work?”, User Context answers “how does this particular person work?” — their tone, their saved workflows, their recent interactions, the accounts they own, the shortcuts they prefer.

Every AI step and chat session receives both layers. Agents ground general reasoning in Business Memory, then personalize the output using User Context — without either layer leaking across users or bleeding into the shared workspace baseline.

What Lives in User Context

Preferences

Preferred tone, writing style, default timezone, language, output format. The user says “I prefer short bullets” once and every future chat and workflow respects it.

Conversation History

Extracted facts from past chat sessions: what the user is working on, which accounts they mentioned, open threads they want to follow up on.

Owned Accounts & Deals

Which companies, contacts, and deals belong to this user. Outreach and reporting workflows scope themselves automatically instead of asking every run.

Saved Workflows & Templates

Personal favorites and frequently-run templates. Useful for chat-first interactions where “run my usual enrichment flow” needs to resolve to something concrete.

Interaction Patterns

Recurring behaviors the system has learned: typical working hours, review cadence, escalation thresholds. Proactive agents use these to time their nudges.

Scoping Model

User Context is stored inside the workspace, but every entry is tagged with the user identifier so reads and writes are transparently scoped. You never have to filter by user manually — the MCP tools and AI step context do it for you based on the calling identity.

Layer	Visibility	Typical Content
Business Memory	All users in workspace	ICP, rubrics, brand voice, catalog
User Context	Single user only	Preferences, owned accounts, history
Workflow Memory	Single workflow only	Per-workflow learnings and counters

Writing User Context

Most User Context is written automatically. Chat extracts preferences from what you say, and workflows write back observations as they run. You can also write directly from an MCP client or an AI step:

store_memory({
  scope: "user",
  key: "preferred_tone",
  value: "direct, no corporate filler",
  category: "preference",
  confidence: 95
})

store_memory({
  scope: "user",
  key: "owned_accounts",
  value: ["Acme Corp", "Globex", "Initech"],
  category: "fact",
  mergeStrategy: "append"
})

scope: "user" ties the entry to the calling user. The same key written by a different user is a different entry.

Reading User Context

Reads follow the same three patterns as Business Memory — direct recall, listing by category, or semantic search — but restricted to the current user's entries:

// Direct recall
recall_memory({ scope: "user", key: "preferred_tone" })

// List all the current user's preferences
list_memories({
  scope: "user",
  category: "preference",
  limit: 50
})

// Semantic search across the user's memories
search_memories({
  scope: "user",
  query: "what accounts am I working on this quarter"
})

Chat sessions automatically inject a compact bundle of User Context alongside Business Memory into the system prompt, so agents have the right personalization without you writing prompts by hand.

How Agents Use Both Layers

When an agent runs on behalf of a user, the memory context is assembled in three phases and passed into the model together:

Phase 1Top Business Memory entries by confidence — the workspace baseline (ICP, rubrics, brand).

Phase 2Semantic search over the Knowledge Graph for entries relevant to the current query, filtered by user visibility.

Phase 3User Context entries for the calling user — preferences, owned accounts, recent patterns.

The combined bundle is capped so it stays compact (roughly a few hundred tokens), keeping the rest of the context window free for the actual task.

Privacy & Isolation

•User Context is never exposed to other users in the workspace. Admins can audit it, but agents running on behalf of user B never see user A's memories.
•User Context stays inside the workspace boundary. It is not shared across workspaces, even if the same person is a member of both.
•You can delete an entry at any time from Settings > Memory, and removing a user from the workspace also removes their User Context.

Next Steps

Business Memory — The shared workspace layer that User Context sits on top of
Persistent Memory — Underlying memory primitive, merge strategies, and TTLs
Proactive Agents — How agents use User Context to time and personalize nudges
Knowledge Graph Overview — Where memories fit in the broader KG architecture