ArtifactFlow

Guide · Choosing a durable home

Where should AI-generated artifacts live?

Keep working context in AI chat, production software in Git, public delivery on a hosting platform, and deliberate reusable outputs in a managed artifact vault when they need stable identity, versions, search, ownership, permissions, and audit without becoming another software project.

Define the artifact-vault layer → Evaluate ArtifactFlow locally →

The deciding question

What job must the storage surface do after generation?

The same HTML dashboard can be temporary chat output, an attachment in a shared folder, the first file in a maintained application, a public web page, or a governed internal artifact. The format alone does not select the right home. Purpose, audience, expected lifetime, change process, and trust boundary do.

AI chat

Best for generation context and short-lived iteration.

Strengths

Prompts, reasoning, alternatives, revisions, and immediate feedback stay together while the work is being produced.

Limits

The durable result is tied to one thread and provider. Ownership, authoritative version, access, taxonomy, and team-wide retrieval are usually unclear.

Keep the conversation when the conversation itself matters. Preserve the selected output elsewhere when people will rely on it after the thread fades.

Downloads and shared drives

Best for familiar file exchange and broad document compatibility.

A shared drive handles ordinary files well and may provide folder permissions, comments, and office-document collaboration. It is often the simplest place for a binary that people already know how to open.

Generated tools can accumulate as ambiguous copies such as final-v3-really-final.html. Executable preview isolation, content-aware search, artifact identity, machine-safe writes, and a consistent audit model are not automatically present.

Git repository

Best when the artifact has become maintained software.

Strengths

Git provides precise source history, branching, code review, automation, dependency management, tests, releases, and collaboration for technical teams.

Limits

A repository, build, deployment, and release process is substantial overhead for every small calculator, runbook, briefing, diagram, or screenshot.

Move an artifact to Git when multiple files, dependencies, production operation, review workflows, or continuous delivery have become part of its real shape.

Static hosting

Best when the primary requirement is delivering a public web page.

Static hosting is excellent for inexpensive, cacheable public delivery. It is not automatically a searchable source-of-truth system with private workspaces, artifact-level permissions, retained source versions, ownership, lifecycle state, and audit.

Serving generated HTML from the same origin as an authenticated control plane also creates a security decision that ordinary hosting does not solve merely by making deployment easy.

Managed artifact vault

Best for reusable internal outputs that deserve governance without a new project.

An artifact vault gives the result a stable record, authoritative payload, retained versions, search, ownership, permissions, safe preview, and audit. People and approved agents can retrieve and improve the same asset through one authority model.

ArtifactFlow occupies this layer for current Markdown, Mermaid, single-file HTML, and normalized image artifacts. It is self-hosted and built for authenticated internal teams.

No public sharing or marketplace

ArtifactFlow's current alpha does not turn private team artifacts into public links, host a public catalog, generate content, or replace a production software delivery platform.

Practical routing rules

Choose the smallest system that honestly matches the artifact's job.

  • Keep it in AI chatwhen it is still disposable working material or its conversational context is the important record.
  • Use a shared file surfacewhen ordinary document collaboration and compatibility matter more than executable preview or artifact-specific governance.
  • Use Gitwhen the output is maintained software with dependencies, review, testing, releases, and operational ownership.
  • Use static hostingwhen public delivery is the primary product and the source-of-truth workflow lives elsewhere.
  • Use an artifact vaultwhen a deliberate internal output must stay findable, versioned, permissioned, safely usable, and available to people and approved agents.
What is an AI artifact vault?The category definition, essential properties, and current ArtifactFlow scope. Artifact identity and versionsHow one stable record accumulates retained immutable content revisions. Safe generated HTMLWhat a preview boundary must do when the retained artifact is executable. MCP for approved agentsScoped search, reads, creates, updates, descriptions, and reverts through the same rules.