ArtifactFlow

Guide · Category definition

What is an AI artifact vault?

An AI artifact vault is a durable system of record for deliberate tools and documents created with AI. It preserves the selected output independently of its original chat and gives that output stable identity, authoritative content, retained versions, search, ownership, permissions, safe use, and audit history.

See the ArtifactFlow lifecycle → Compare storage choices →

The missing layer

The useful result should outlive the conversation that produced it.

AI chats are effective working spaces, but a finished calculator, briefing, diagram, screenshot, runbook, or one-file dashboard may need a different life. People must know which copy is authoritative, who owns it, who can use it, what changed, and whether executable content can be opened without inheriting the trust of the application around it.

A vault preserves the output, not the conversation. It does not try to become agent memory, a vector database, a chat archive, an AI generator, or a general software deployment platform.

Essential properties

A folder of generated files is not yet an artifact vault.

  • Stable identityOne artifact keeps the same record even as its title, description, tags, lifecycle state, or content changes.
  • Authoritative payloadThe system retains the source or normalized original that people and approved tools can retrieve again.
  • Immutable content versionsEach retained content change creates a new revision instead of silently overwriting history.
  • Search and ownershipArtifacts stay findable and accountable outside the chat or laptop where they began.
  • Permissions and auditReads and writes pass through one authority model and important changes remain explainable.
  • Format-aware safe useMarkdown, executable HTML, images, and future documents need different preview and processing boundaries.

Identity and change

Content history and ordinary metadata are different kinds of change.

In ArtifactFlow, a page has a stable identity and a current version pointer. Retained source changes create immutable versions; metadata and draft state can change without manufacturing a content version. Reverting creates a new version copied from an older payload, so history remains append-only.

Inspect the implementation contract

The repository's artifact lifecycle document defines current invariants, format payloads, draft behavior, version creation, and roadmap boundaries.

Safe consumption

Some artifacts are documents. Some are executable programs.

Retaining an HTML file is easy. Running untrusted HTML near authenticated sessions is the hard part. A credible artifact vault treats preview as a trust-boundary problem, not a rendering convenience. ArtifactFlow serves generated HTML from a separate cookieless origin inside a sandboxed frame with restrictive response policy and short-lived signed access.

Those layers reduce risk but do not erase it. Scanning is advisory, and the documented self-navigation and browser-dependent WebRTC residuals still matter. The safe generated HTML guide explains the boundary.

ArtifactFlow today

The public alpha implements a narrow, testable slice.

ArtifactFlow currently supports Markdown pages with Mermaid diagrams, self-contained single-file HTML artifacts, and normalized PNG/JPEG images in personal or shared workspaces. Approved AI clients can search and retrieve authoritative content, create supported artifacts, update editable descriptions, and preserve new text or HTML versions through scoped MCP operations.

Searchable PDF and Word artifacts are the next product focus. Public sharing, a marketplace, AI generation, general static hosting, and full Confluence parity are not part of the current product boundary.

Artifact workflowCurrent identity, version, draft, metadata, and format invariants. Security architectureMaintained controls, threat boundaries, advisory safeguards, and residual risks. MCP for approved agentsScoped search, reads, creates, updates, descriptions, and reverts. Where should artifacts live?When chat, a shared drive, Git, static hosting, or a vault is the right home.