01
Preserve the working system first
Freeze the known-good environment before you rebuild it. Record the Flowise version, runtime, Docker tag when applicable, database version, model identifiers, custom components and deployment configuration.
- Back up the database and flow definitions.
- Export portable configuration where your deployment supports it.
- Keep secrets separate and record only where they are stored and how they are rotated.
02
Map hidden dependencies
The visible graph is only part of the system. Inventory models, vector stores, files, APIs, OAuth applications, webhooks, retry rules, schedules, queues, memory, state and idempotency requirements.
- Models and tool schemas can behave differently on another platform.
- OAuth callback URLs and credentials often need to be recreated.
- State and side effects require explicit migration tests.
03
Choose a target by workload
Langflow is a natural evaluation target for visual agent construction, n8n for broader cross-application automation with AI steps, and Dify for a broader AI-application and plugin architecture. None should be assumed to import Flowise JSON as a one-click compatible migration.
- Compare the actual capabilities your production flow uses.
- Build a node/capability migration matrix.
- Keep the model constant during platform tests when possible so you know what changed.
04
Test before cutover
Build a golden test set containing normal requests, edge cases, tool failures, expired credentials, rate limits and repeated requests that exercise idempotency. Parallel-test the replacement before moving all traffic.
- Measure correctness, latency, tool reliability and cost.
- Keep the old deployment restorable during a rollback window.
- After cutover, revoke obsolete credentials and remove abandoned public endpoints.
Sources
Primary and supporting sources
Facts were rechecked against the linked sources immediately before publication. Pricing, product availability and rollout status can change.