Skip to main content

Workspaces and Knowledge

Workspaces organize people, agents, documents, tasks, workflows, and knowledge under one operational boundary — typically one per part of your business: a property you manage, a support inbox, a marketing function.

Workspaces​

A workspace scopes:

  • Agents and skills (via service mappings and subscriptions).
  • Conversations and tasks.
  • Goals and their execution plans.
  • Documents and file assets.
  • Knowledge sources.
  • Integration credentials and channels.
  • Governance rules, permissions, and audit history.

A workspace page has tabs for its overview (services, matched agents, current work, operating score), goals, tasks, documents, channels, agents, activity, and settings. Workspace chat is contextual: it sees the workspace's services, goals, agents, and recent activity, and can start workflows directly from a message.

Creating a workspace is a guided, chat-style flow — Manor helps you fill in the workspace's purpose, services, staff, channels, and seed knowledge, then creates it. You can also install a blueprint to start from a packaged configuration.

Knowledge​

The knowledge system indexes documents so agents can cite them. Upload PDFs, Word and Excel files, Markdown, plain text, slide decks, or images to Knowledge, and group documents into knowledge bases by purpose ("House Rules", "Vendor Contacts", "FAQ"). Attach groups to a workspace and its agents use them automatically.

How Retrieval Works​

Retrieval is chunk-level and hybrid, fused with Reciprocal Rank Fusion:

  1. Documents are split into ~1,600-character chunks; each chunk gets its own embedding (pgvector, generated by the bundled Ollama service).
  2. A query runs vector search and trigram/lexical search (CJK-safe) over chunks, plus filename matching.
  3. Results are deduplicated to the best chunk per document and rank-fused, so exact-term matches and semantic matches both surface.

Freshly uploaded documents that aren't indexed yet can still be found in explicitly scoped searches — the runtime falls back to reading the file directly within a workspace or named knowledge-base scope.

The search bar (and GET /api/v1/search?q=) does fast substring search across tasks, documents, agents, and conversations, honoring per-user document and workspace permissions. It complements — not replaces — the semantic rag retrieval agents use for content questions.

Data Isolation​

Everything is scoped to your entity (organization), and workspace boundaries apply within it: agents see the workspaces they're mapped into, and workspace-scoped results are permission-filtered per user. Keep workspace boundaries meaningful — do not mix unrelated teams or clients in one workspace unless they intentionally share agents, documents, and memory.