Your Obsidian vault may hold years of decisions, project notes, diagrams, and research. Moving these files into an AI-connected workspace sounds simple because the notes are Markdown. But an import can appear successful while quietly losing the relationships that make the knowledge useful: a wiki link stops resolving, an image points to an old directory, or an assistant treats a speculative draft as an approved decision.
An Obsidian-to-AI-memory migration succeeds when your knowledge remains readable, its limits are documented, and the right assistants can retrieve the right evidence. This checklist uses a reversible pilot before a whole-vault move.
1. Separate Markdown portability from feature compatibility
Markdown portability is a starting point, not a guarantee that two apps behave identically. Obsidian can rely on wiki links, embedded files, block references, community plugins, custom properties, themes, and application-specific views.
Make two inventories:
- Information to retain: note bodies, relative paths, source documents, attachments, dates, decision context, and links.
- Behavior to reassess: plugin queries, custom callouts, templates, Canvas layouts, and automated workflows.
For example, `[[Architecture Decision 17]]` can remain readable Markdown while no longer resolving to the intended note. Preserving a Dataview query as text does not prove another app can execute it. Treat unsupported behavior as a known limitation rather than a successful migration.
2. Preserve the original and choose a representative pilot
Create a separate, dated backup of the full Obsidian folder, including attachments. Keep it outside the working import folder. Never experiment on the only copy.
Choose a representative project of roughly 20–50 notes, including nested folders, wiki links, a PDF or image, frontmatter, a recorded decision, and a plugin-dependent note. This is a suggested pilot size, not a product limit.
Record actual file and attachment counts, critical relationships, and exceptions. An example manifest:
```yaml
pilot: launch-planning
source_copy: obsidian-backup
notes_expected: 36
attachments_expected: 11
critical_links:
- "Decision 17 -> Launch Checklist"
- "Meeting 04 -> Pricing Decision"
manual_review: - dataview-query
- canvas-layout
owner: project-maintainer
```
Replace the example numbers with your own. A baseline inventory turns later claims of success into something you can verify.
3. Audit broken links before importing
Start with Phasoric's free Vault Audit, or a comparable local Markdown checker. The public audit inspects a selected folder in the browser without uploading its note content or initiating sync. Review unresolved links, orphan notes, duplicates, attachments, and migration warnings.
Separate existing problems from new migration defects. If a link was already broken in Obsidian, its failure after import is not evidence that the importer broke it. Conversely, a link that worked before and fails afterward demands investigation.
For each critical document, record the source path, expected destination path, link target, attachment accessibility, and disposition. Explicitly resolve ambiguous links between two notes with the same title.
4. Define acceptance rules for each construct
Before importing, decide what success looks like for the actual content in your pilot:
| Content type | Acceptance test |
|---|---|
| Plain Markdown | Text, lists, headings, and code blocks remain readable |
| Wiki links and embeds | Each sampled target opens the intended note or file |
| Relative attachments | Image and PDF binaries open from the new note |
| YAML properties | Required values retain their intended meaning |
| Plugin queries and views | Limitations are flagged or rebuilt manually |
| Decisions and drafts | Historical status survives; drafts are not treated as approved |
A file being present is not proof its functionality survived. For example, preserving an attachment filename does not establish that the attachment itself was copied.
5. Preview the Phasoric import and collision rules
Phasoric documents a migration workflow for Obsidian, ENEX, Markdown folders, and supported documents. The application's Settings → Migration wizard provides a preflight covering links, attachments, property parsing, exact-content duplicates, path collisions, and warnings.
Review that report before making destination changes. Choose distinct policies for path collisions and exact-content duplicates. Notes sharing a path need not have the same content; different paths may contain identical files.
The documented implementation uses bounded commits with recoverable checkpoints. If a tab closes or an import fails, review the stored session and use Resume instead of blindly restarting the entire batch. The report records planned checks and realized results. This is a recovery mechanism, not a guarantee against all storage failures.
Source files remain untouched. A pre-existing remote path can require a separate reviewed cloud-sync confirmation. Avoid assuming that a successful local import automatically updates hosted storage.
6. Verify relationships with real questions
Open a migrated decision note. Follow at least five representative links, including one renamed or ambiguous target. Open a linked image and a PDF. Search for a phrase buried in a nested note. Compare complex frontmatter against its source.
Then ask a real project question: Why did we change the launch date? Locate the source meeting, decision record, and any superseding decision. An answer without inspectable evidence is not adequate verification.
Log each discrepancy with source path, destination path, observed failure, owner, and disposition. Useful dispositions include fix link, manual rebuild, exclude from agent access, and accepted limitation.
Do not expand the migration while critical links or source documents remain unresolved.
7. Preserve provenance and decision status
An imported document can be authentic without being current or approved. Phasoric's migration guide says newly imported knowledge enters its trusted knowledge ledger as Draft, not automatically Verified. Replacing already tracked material marks the trust record Changed while preserving previous fingerprints.
That distinction matters when old plans and new decisions coexist. Keep abandoned options where historically useful, but label their status, effective date, and owner. A filename containing “final” is not evidence of ongoing authority.
For important records, create or confirm a decision summary with source links and a review date. The related guide on preserving architecture decisions across AI sessions shows how to keep the rationale attached.
8. Connect AI only after validating storage and permissions
Importing a vault and exposing it to ChatGPT or Claude are separate steps. A local-only vault does not automatically become reachable through remote MCP. Hosted MCP access depends on authentication, a supported plan, and an explicitly authorized hosted mapping.
Start with a restricted test collection. Ask an assistant for one permitted note with its source. Then request an excluded note and confirm it is not returned. When local-only material is unavailable remotely, inspect reachability and mapping rather than trying a more forceful prompt.
For deployment tradeoffs, see local versus hosted AI memory and the public MCP contract. A successful import is not approval to disclose or modify its contents.
9. Six acceptance tests before scaling up
Document each result, not just a pass/fail label:
- Inventory: note and attachment counts reconcile, with skips explained.
- Links: sampled wiki links, ordinary links, and embeds reach the correct targets.
- Properties: critical dates, identifiers, and decision statuses survive or have a repair plan.
- Compatibility: plugin-dependent workflows are supported, rebuilt, or explicitly unavailable.
- Trust: unreviewed imports are not passed off as verified decisions.
- Access: an authorized assistant retrieves a permitted source without reaching excluded material.
Keep the evidence next to your migration report. If a critical case fails, repair the pilot before loading more folders.
Keep the working system until the new one proves itself
Moving from Obsidian should not require abandoning useful plugins or pretending that every feature has an equivalent. You can keep Obsidian while evaluating an AI-ready Markdown workspace.
The outcome that matters is a traceable path from original files to readable notes, reviewed decisions, and appropriately scoped AI retrieval. Run the local vault audit, migrate a bounded sample, and expand only when the tests pass.
Frequently asked questions
Will an Obsidian vault migration preserve every plugin and dashboard?
No. Core Markdown is portable, but community plugins, queries, Canvas layouts, themes, and application-specific behavior require separate review or rebuilding. Use the preflight report to identify limitations.
Does importing an Obsidian vault automatically expose it to ChatGPT or Claude?
No. Import and remote AI access are separate. Hosted MCP requires a supported plan, authorization, and an explicitly configured hosted mapping. Local-only notes are not automatically reachable.
Can I test a migration without modifying the original vault?
Yes. Preserve a source copy, run a local vault audit, and use the migration preview. The source remains untouched while you validate the results.