Busabase × Claude.aiKnowledge Base for Claude.ai
Claude Projects can hold shared files, instructions, conversations, and project-specific memory. Claude can also create artifacts and act through connectors. Busabase gives selected outcomes a shared home where every change is a Change Request: attributed, diffed, and reversible, with review when you choose.
Six context layers
A project can contain useful knowledge without becoming the system of record.
Each Claude surface has a clear job. Keeping those jobs separate prevents convenient context from quietly acquiring organizational authority.
| Layer | Primary job | Authority boundary |
|---|---|---|
| Project knowledge | Files and references available across project chats | Context supplied by project editors |
| Project instructions | Persistent response guidance within the project | Behavior, not changing facts |
| Project memory | Continuity and summary isolated to one project | User- and org-controlled memory |
| Chat search | Retrieve cited details from eligible conversations | Conversation provenance |
| Artifacts | Self-contained outputs for viewing, editing, or sharing | Deliverable, not automatic approval |
| Busabase record | Shared claim with owner, lifecycle, and history | Every edit is a Change Request; review is optional per key |
Project knowledge
Uploaded material improves answers; it does not certify the material as current.
Claude Projects create self-contained workspaces with their own chats, files, knowledge base, and instructions. Team and Enterprise projects can be shared with view or edit permissions. Preserve source ownership and review dates when project material informs a durable claim.
Who supplied the file or text?
Which project and audience can use it?
When was it last verified?
What exact conclusion is being promoted?
Memory and chat search
Continuity is personalized retrieval, not a shared team ledger.
Claude stores memory as individual topics and keeps project memory separate from non-project chats. Chat search retrieves within defined conversation boundaries and shows citations. Users and organizations can pause, edit, reset, or disable memory. Promote stable team facts independently so a personal memory change cannot silently rewrite shared state.
Roles, preferences, recurring context
Project details and ongoing work
User-managed topics
Evidence-backed claim
Named accountable owner
Effective and next-review dates
Artifacts
A polished artifact is a deliverable, not proof that its inputs or conclusions were approved.
Artifacts are significant self-contained outputs that can be edited, shared, or connected to external services. Preserve the artifact link or export as evidence, then review the underlying claim, source coverage, and intended use separately.
Connector boundary
Source access and action permission do not make connector output canonical.
Claude inherits each person's permissions from connected services. Organization controls may further mark tools Always allow, Needs approval, or Blocked. Custom remote MCP connectors run from Anthropic's cloud and must be publicly reachable or use documented private-network options.
Can this identity read or modify the source system?
Is this connector action allowed now?
Who wrote the result, what changed, and can it be rolled back?
Three different decisions
Sharing, executing, and accepting a claim must remain separate.
A project editor can change project knowledge. A connector can execute within source-system permission. Neither role necessarily owns the business decision represented by the result.
Knowledge promotion contract
Store a concise claim that can survive outside Claude's project and memory lifecycle.
Keep chat, uploaded files, and artifact content in their native systems. The record connects those artifacts to a current, versioned state.
knowledge_keyStable identity across chats and projectsproject_refProject and relevant knowledge sourcechat_refsCited conversations or research sessionsartifact_refOutput revision used as evidenceconnector_evidenceSource record, tool call, and resulting stateclaimConcise statement proposed as currentownerPerson accountable for acceptancefreshnessObserved, effective, and review datesdecisionAccepted, conditional, rejected, or supersededPractical questions
Claude.ai knowledge base FAQ
Does Busabase replace Claude Projects?
No. Projects organize chats, instructions, files, and project context. Busabase stores selected claims that must be shared beyond one project, each with its change history.
Is project memory visible across all projects?
No. Anthropic documents separate memory and summaries for each project.
Is an Artifact automatically an approved asset?
No. It is a generated deliverable. Approval depends on evidence, ownership, and intended use.
Does enabling a Connector grant access to every source record?
No. Claude inherits the connected person's source-system permissions, and organization tool policy can narrow actions further.
Should connector results be written directly as canonical facts?
They can be: with a write-level key the Change Request merges immediately and stays attributed and reversible. To check first, give Claude a proposal-only key so its writes wait in the Inbox, and include source evidence and an owner.
Start with one project outcome
Keep rich context in Claude and promote only the claim that needs shared trust.
Choose one artifact or connector result with clear sources. Attach its project, evidence, owner, and freshness rule, and write it back as a Change Request, through a proposal-only key if you want to review it before another team or agent relies on it.