Skip to content

FAQ

MIFY is a unified AI workflow orchestration platform. You build workflows visually, connect to 140+ AI providers, and run with full control including human approval gates and audit trails.

MIFY stands for “Meta-Intelligent Flow Yield.”

MIFY supports local deployment with open-source AI models via Ollama. The platform runs locally with no cloud dependency required.

  • Node.js 22+
  • PostgreSQL 15+ with pgvector extension
  • pnpm 9+ package manager

For AI capabilities, install Ollama locally or add API keys for cloud providers.

ServicePort
Web UI3006
API Server3010
Home Runtime WebSocket3020

Go to Settings → Credentials. Select your provider (OpenAI, Anthropic, etc.) and enter your API key. Keys are encrypted at rest.

126 native nodes across 24 categories including LLM, RAG, Transform, Logic, Platform, Agent, Safety, Sandbox, and Enterprise nodes.

Can I import workflows from other platforms?

Section titled “Can I import workflows from other platforms?”

Yes. MIFY imports from Flowise, Dify, Langflow, and n8n. Go to Canvas → Import and paste the workflow JSON/YAML.

HITL lets you add approval gates to workflows. Execution pauses at a Pause node until a human approves or rejects. Pending approvals appear in your Inbox.

Yes. Every published workflow is captured as a GraphVersion snapshot. Use /graphs/[graphId]/versions to publish, archive, restore, and diff (line-level via diff-match-patch) versions. See Execution & Trust.

What multi-agent strategies does MIFY support?

Section titled “What multi-agent strategies does MIFY support?”

Four strategies, all wired through the MultiAgentNode:

  • Plan-and-Execute — decompose goal then execute subtasks
  • Reflection — generate-then-critique with self-correction
  • Supervisor — hierarchical coordinator + specialized workers
  • Swarm — peer-to-peer agent network with message bus

See the Multi-Agent node reference.

What’s the Cloudflare Workers AI fallback?

Section titled “What’s the Cloudflare Workers AI fallback?”

If your primary LLM provider goes down or hits a quota, MIFY can fall back to Cloudflare Workers AI automatically. Three-stage fallback (flat settings + kill switch + routing). Admin emergency disable lives behind the gateway page (API: /api/admin/cf-killswitch/*). See AI Provider System.

Yes — MIFY’s Backend Capability Registry gives every workspace a per-capability setting:

CapabilityWorkspace Setting Page
LLM/workspaces/[id]/settings/llm
Document Parser/workspaces/[id]/settings/parser
Browser/workspaces/[id]/settings/browser
Sandbox/workspaces/[id]/settings/sandbox

Each capability also has an admin fallback page at /admin/{llm,parser,browser,sandbox}-fallback for org-wide defaults.

Navigate to /content-gen and select a workflow. MIFY supports three paths:

  • MIFY Queue — GPU-powered generation via ComfyUI workflows
  • AI Provider — Direct calls (OpenAI DALL-E, Cloudflare, etc.)
  • ComfyUI Local — Run on your local GPU

Yes. Go to /training/new, upload a dataset, select a training preset, and submit. Trained models are stored in the model registry.

When you buy an asset, funds are held in escrow until the seller transfers the asset and you verify the transfer. Disputes are handled through the platform.

Sellers can be Basic (email verified), Enhanced (identity + financial), or Premium (full business verification). Higher tiers get trust badges and better visibility.

A local runtime that runs AI coding agents (Claude Code, Codex) on your machine with human review gates. No cloud required.

Yes. Create a cron schedule with a task prompt and the agents will run automatically at the specified times.

Yes — two embed types are available:

  • Public Chat Widget (Demo 0) — a no-login AI chat assistant your visitors can use without a MIFY account. POST to /api/chat-widget/message. Hardened with rate-limit, quota guard, and redaction. Embed code generated at /settings/embed.
  • Forms — multi-page form submissions trigger any workflow. Drag-drop builder at /forms/[formId]/builder; render at /forms/[formId]. The FormTriggerPlugin handles the trigger side.
  • Public Demo Runner — wrap any starter template at /templates/starter/[slug] so visitors can run it without an account (rate-limited globally).

When an embedded MIFY widget needs to authenticate the visitor against their MIFY account, the Site Consent System issues a one-time challenge → reference token → session, with full GDPR audit trail. Cloudflare Turnstile protects the consent grant step. See Franchise Operators for details.

Can multiple users edit the same workflow at the same time?

Section titled “Can multiple users edit the same workflow at the same time?”

Yes. The canvas runs on Hocuspocus + Yjs — multiple users see each other’s edits in real time, with live presence cursors showing who’s on the workflow. Auth is enforced via the SuperTokens session handshake. Y.Doc state is persisted in Postgres so connections can reattach without losing work.

What’s the difference between durable and regular execution?

Section titled “What’s the difference between durable and regular execution?”

Regular (SimpleExecutor) — runs in-process in the API server. Fast, low overhead, but state is lost if the process restarts.

Durable (Temporal) — each node runs as a Temporal Activity. The run survives process restarts, scales horizontally, and produces a full event history that lets you replay or fork from any prior step. Switch a workflow to durable in /graphs/[graphId]/settings.

Can I replay a previous run from a specific step?

Section titled “Can I replay a previous run from a specific step?”

Yes, for durable runs. Open /workflows/[graphId]/runs/[runId], scrub to the step you want, and click “Replay from here” or “Fork with overridden inputs.” API: POST /api/temporal-time-travel/replay and /fork.

When I edit a workflow, does it re-run everything?

Section titled “When I edit a workflow, does it re-run everything?”

No. Partial execution (W16-9) re-runs only the dirty nodes — those you changed plus their downstream dependents. Unchanged subgraphs reuse cached outputs from PartialExecutionCache. The canvas shows a DirtyNodesIndicator on every node that will re-execute.

How do I call a workflow from outside MIFY?

Section titled “How do I call a workflow from outside MIFY?”

Publish the workflow, then POST to /api/workflow-endpoints/:slug with an API key. Each Published workflow has:

  • A stable HTTP endpoint with per-API-key rate limiting
  • An auto-generated OpenAPI 3.1 spec at /api/workflow-endpoints/:slug/openapi.json
  • A sample curl + endpoint info panel on /graphs/[graphId]/versions

Can other agent frameworks call my workflows?

Section titled “Can other agent frameworks call my workflows?”

Yes. Every Published workflow is auto-exposed across three agent protocols:

  • MCP (Model Context Protocol) — /api/workflow-mcp/list_tools + call_tool. Works with Claude Desktop, Cursor, and any MCP client.
  • A2A v1.0/api/a2a-server/* with auto-populated AgentCard
  • LangGraph Agent Protocol/api/agent-protocol/* (bidirectional)

All three live in the @mify/agents-network package.