--- title: "The Complete Guide to Leaving Vendor AI Memory" description: "Move AI work out of ChatGPT, Claude, Notion, or an agency archive into readable files you own, can test, and can hand to the next operator." canonical: "https://scalewithsearch.com/articles/leaving-vendor-ai-memory-guide" date: "2026-08-21" modified: "2026-09-19" --- ## Site navigation - [Scale With Search](https://scalewithsearch.com/) - Real estate - Real estate - [Real estate](https://scalewithsearch.com/for/real-estate) - Work - Start here - [Send your brief](https://scalewithsearch.com/work#send-your-brief) - [Prepare your six-question brief](https://scalewithsearch.com/work#prepare-your-six-question-brief) - Build - [Site build, content library with SEO, signal desk](https://scalewithsearch.com/work) - For your business - Trades and home services - [Auto body and collision shops](https://scalewithsearch.com/for/auto-body-and-collision-shops) - [Foundation and home repair contractors](https://scalewithsearch.com/for/foundation-and-home-repair) - [Garage door and fencing contractors](https://scalewithsearch.com/for/garage-door-and-fencing-contractors) - [HVAC contractors](https://scalewithsearch.com/for/hvac-contractors) - [Janitorial and commercial cleaning companies](https://scalewithsearch.com/for/janitorial-and-commercial-cleaning) - [Locksmiths](https://scalewithsearch.com/for/locksmiths) - [Moving companies](https://scalewithsearch.com/for/moving-companies) - [Pest control companies](https://scalewithsearch.com/for/pest-control-companies) - [Plumbing and electrical contractors](https://scalewithsearch.com/for/plumbing-and-electrical-contractors) - [Restoration and water or fire damage companies](https://scalewithsearch.com/for/restoration-and-water-fire-damage) - [Roofing companies](https://scalewithsearch.com/for/roofing-companies) - [Towing companies](https://scalewithsearch.com/for/towing-companies) - [Tree services and landscaping companies](https://scalewithsearch.com/for/tree-services-and-landscaping) - [Solar installers](https://scalewithsearch.com/for/solar-installation) - [General contractors](https://scalewithsearch.com/for/general-contractors-and-construction) - [Paving, concrete, and flooring contractors](https://scalewithsearch.com/for/paving) - Practices and professional services - [Bookkeeping and tax practices](https://scalewithsearch.com/for/bookkeeping-and-tax-practices) - [Dental practices](https://scalewithsearch.com/for/dental-practices) - [Family and criminal defense law firms](https://scalewithsearch.com/for/family-and-criminal-defense-law-firms) - [Med spas and aesthetics practices](https://scalewithsearch.com/for/med-spas-and-aesthetics) - [Personal injury law firms](https://scalewithsearch.com/for/personal-injury-law-firms) - [Veterinary clinics](https://scalewithsearch.com/for/veterinary-clinics) - [Gyms and fitness studios](https://scalewithsearch.com/for/fitness) - [Therapy and outpatient health practices](https://scalewithsearch.com/for/therapy-and-outpatient-health) - [Medical billing companies](https://scalewithsearch.com/for/medical-billing) - [Insurance agencies](https://scalewithsearch.com/for/insurance-agencies) - [Financial advisors](https://scalewithsearch.com/for/financial-advisors) - [Property management companies](https://scalewithsearch.com/for/property-management) - [Recruiting and staffing agencies](https://scalewithsearch.com/for/recruiting-and-staffing) - [Architects and interior designers](https://scalewithsearch.com/for/architects-and-interior-designers) - [Logistics and supply chain companies](https://scalewithsearch.com/for/logistics-and-supply-chain) - Agencies, MSPs, and manufacturing - [IT and managed service providers](https://scalewithsearch.com/for/it-and-managed-service-providers) - [Machine shops and precision manufacturers](https://scalewithsearch.com/for/machine-shops-and-precision-manufacturing) - [Marketing agencies and freelancers](https://scalewithsearch.com/for/marketing-agencies-and-freelancers) - [SEO agencies and consultants](https://scalewithsearch.com/for/seo-agencies-and-consultants) - [Small manufacturers and fabricators](https://scalewithsearch.com/for/small-manufacturers-and-fabricators) - Restaurants, shops, studios, and nonprofits - [Restaurants and hospitality businesses](https://scalewithsearch.com/for/restaurants-and-hospitality) - [Retail stores and ecommerce sellers](https://scalewithsearch.com/for/retail-and-ecommerce) - [Photographers, event planners, and travel agents](https://scalewithsearch.com/for/photographers) - [Churches and nonprofits](https://scalewithsearch.com/for/churches-and-nonprofits) - [All industries](https://scalewithsearch.com/for/) - Learn - For your office - [Office job guides](https://scalewithsearch.com/guides/) - [Browser calculators](https://scalewithsearch.com/tools/) - Start here - [How it works](https://scalewithsearch.com/how-it-works) - [Free Starter Kit](https://scalewithsearch.com/kit/business-memory-starter-kit.zip) - [Synthetic specimen](https://scalewithsearch.com/specimen/working-session-specimen.zip) - Guides - [The Complete Guide to Business Memory for AI Agents](https://scalewithsearch.com/articles/business-memory-for-ai-agents-guide) - [The Complete Small-Business Guide to AI Agent Governance](https://scalewithsearch.com/articles/ai-agent-governance-guide-small-business) - [The Complete Guide to Leaving Vendor AI Memory](https://scalewithsearch.com/articles/leaving-vendor-ai-memory-guide) - Articles by cluster - [Business memory](https://scalewithsearch.com/articles/business-memory-for-ai-agents-guide) - [Agent governance](https://scalewithsearch.com/articles/ai-agent-governance-guide-small-business) - [Migration and ownership](https://scalewithsearch.com/articles/leaving-vendor-ai-memory-guide) - For machines - [llms.txt](https://scalewithsearch.com/llms.txt) - [llms-full.txt](https://scalewithsearch.com/llms-full.txt) - [Machine view](https://scalewithsearch.com/?view=machine) - Company - Evidence - [Proof](https://scalewithsearch.com/proof) - Company - [About](https://scalewithsearch.com/about) # The Complete Guide to Leaving Vendor AI Memory. The agency relationship ends on Friday. On Monday, the owner learns that three years of prompts, accepted drafts, customer exceptions, and reporting notes live inside the agency's Notion workspace and two employees' ChatGPT accounts. A shared ZIP arrives, but nobody can tell which offer is current, which client note was corrected, or how the monthly report was produced. The files were delivered. The operation was not transferred. Leaving vendor AI memory means moving durable business work out of a provider account or agency archive into files the business can read, review, back up, test, and hand to another operator. The goal is not to recreate every conversation. The goal is to preserve evidence, promote current records, rebuild one recurring job, and prove that the next authorized worker can run it without the old account. This method applies to ChatGPT, Claude, Notion, agency portals, shared drives, and mixed archives. Product export formats differ. The custody, classification, promotion, and acceptance steps remain stable. ## Start at your exit stage - **No exports yet:** define the exit and preserve raw copies before changing access. - **Exports received:** inventory, reconcile coverage, and classify before promoting anything. - **Deadline approaching:** protect access, rebuild the survival job, and test the handoff before cancellation. Follow the stages: 1. [Define the exit](#define-the-exit-before-exporting) 2. [Preserve raw exports](#preserve-the-raw-exports) 3. [Inventory the archive](#build-an-inventory-before-reading-everything) 4. [Classify records](#classify-before-promoting) 5. [Rebuild one job](#rebuild-one-recurring-job) 6. [Test acceptance](#build-the-acceptance-test-before-cutting-access) 7. [Rehearse the handoff](#rehearse-the-handoff-with-a-person-who-did-not-build-it) 8. [Record the result](#record-the-final-handoff) ## Define the exit before exporting An export is a collection of data. An exit is an accepted operating state. Write the finish line first. Name the archive in scope, the business owner, the recurring job that must survive, the required records, the allowed exclusions, the test, and the stopping boundary. ```yaml # migration/exit-brief.yaml migration_id: vendor-exit-2026-08 owner: operations director sources: - ChatGPT account export - Claude account export - Notion workspace export - agency final archive survival_job: monthly-client-report required_result: - immutable raw copies - readable inventory - reviewed current records - task brief and source order - acceptance fixture - runbook and receipts stopping_point: independent draft and handoff receipt must_not: - delete provider data - close accounts - rotate credentials - publish or send a report ``` This prevents the archive size from becoming the project definition. The broader guide to [moving AI work out of a vendor archive](/articles/move-ai-work-out-of-vendor-archive) uses one recurring job as the unit of migration. The article on [what a working session returns](/articles/what-a-working-session-returns) shows the smaller result when the owner needs one job worked through, with an owned context path and one tested output, rather than a full transfer. ## Preserve the raw exports Request exports through the official account or workspace controls. Keep the original packages unchanged. Record who requested each export, the account, request time, delivery time, filename, size, and cryptographic hash. OpenAI documents a data export through ChatGPT's Data Controls and Privacy Portal. The export includes chat history and other account data, with availability and delivery rules defined by the provider. [OpenAI: How do I export my ChatGPT history and data?](https://help.openai.com/en/articles/7260999-how-do-i-export-my-chatgpt-history-and-data) Anthropic documents exporting Claude data from account settings, with organization controls and product-specific availability. [Anthropic: Export your Claude data](https://support.claude.com/en/articles/9450526-export-your-claude-data) Notion documents workspace export options for HTML, Markdown and CSV, and PDF, subject to workspace settings and plan limits. [Notion: Export your content](https://www.notion.com/help/export-your-content) Check the delivery limits before you schedule the exit. As checked September 19, 2026, OpenAI says an export can take up to 7 days to arrive, and the ChatGPT and Claude download links expire 24 hours after delivery. Anthropic limits organization exports to the Primary Owner and does not offer export in its mobile apps. Notion says a workspace export excludes pages the exporter cannot access, such as other users' private pages, so an agency workspace export can omit private material. Do not claim completeness from a successful download. Record what the provider says the export covers, then compare it with the expected accounts, workspaces, dates, attachments, and databases. ```yaml # migration/raw/export-register.yaml exports: - source: chatgpt account: owner@example.com requested_at: 2026-08-21T09:00:00-04:00 received_at: 2026-08-21T09:18:00-04:00 file: raw/chatgpt/export-2026-08-21.zip bytes: 18422091 sha256: 3bd6... coverage_claim: provider account export verified_open: true - source: notion workspace: agency-client-space file: raw/notion/export-2026-08-21.zip sha256: 81ac... coverage_claim: workspace export selected by administrator verified_open: true ``` Store raw files read-only where practical. Do not normalize in place. Conversion errors and later disputes require an unchanged baseline. ## Build an inventory before reading everything Inventory files and conversations by metadata first. Capture source system, original identifier, title, date range, participants when known, attachment count, file type, business or client scope, and review state. Do not use generated summaries as the only inventory. Keep stable links back to raw items. A reviewer should be able to move from a promoted decision to the original conversation or document. ```csv source,item_id,title,created,modified,scope,type,attachments,review_state,raw_path chatgpt,conv-4481,Client North renewal,2026-03-04,2026-04-03,client-north,conversation,2,unreviewed,raw/chatgpt/conversations.json notion,page-182,Monthly report SOP,2025-11-10,2026-07-15,reporting,page,4,unreviewed,raw/notion/Monthly report SOP.md agency,file-991,Offer notes,2026-01-02,2026-06-11,company,document,0,unreviewed,raw/agency/offer-notes.docx ``` Count inventory records by source and compare them with visible account or workspace counts when the provider exposes those counts. Note gaps instead of filling them with assumptions. The inventory is a map, not business memory. It helps review and deduplication. Authority is assigned later. ## Convert into a readable archive Create a readable derivative layer without overwriting the raw layer. Preserve titles, timestamps, speakers where unambiguous, source IDs, and attachment references. Keep uncertain authorship labeled as uncertain. For ChatGPT exports, the guide to [converting a ChatGPT export to Markdown](/articles/chatgpt-export-to-markdown-business-memory) separates the immutable JSON or HTML package from readable conversation files and promoted records. The broader [ChatGPT-history migration method](/articles/migrate-chatgpt-history-to-business-memory) covers inventory, classification, review, and the first cold run. Use a conversion manifest: ```yaml # migration/converted/conversion-receipt.yaml tool: convert-chat-export tool_version: 1.3.0 input: raw/chatgpt/export-2026-08-21.zip input_sha256: 3bd6... output_root: converted/chatgpt/ conversations_seen: 428 conversations_written: 428 attachments_referenced: 91 parse_errors: 0 speaker_uncertain: 7 completed_at: 2026-08-21T11:42:00-04:00 ``` The receipt makes conversion repeatable. It also identifies records needing manual review. Do not assign a named speaker when an export lacks reliable turn attribution. Plain Markdown, text, CSV, JSON, and standard image or document formats make a useful durable layer when they preserve required structure. [Why plain text is useful for AI memory](/articles/plain-text-ai-memory) explains the readability and portability benefits without claiming every business record belongs in one text file. ## Classify before promoting Every archive item starts as evidence. Review candidates can move into one of three states: keep as a current record, reject as noise or superseded material, or hold for an authorized reviewer. Keep items that carry a confirmed current fact, accepted decision, stable procedure, approved example, accepted correction, or unresolved work item. Reject duplicates, generic model advice, abandoned drafts, obsolete instructions, and claims that cannot be traced to a qualifying source. Route ambiguity, sensitivity, and conflicting authority to review. The guide to [what to keep, reject, or review from a chat archive](/articles/what-to-keep-from-ai-chat-archive) provides the triage fields. It protects against two opposite errors: discarding useful evidence and promoting every polished answer into policy. ```yaml # migration/review/REV-000184.yaml candidate_id: REV-000184 source_item: chatgpt:conv-4481 source_locator: message-19 proposed_record: Client North renews on 2026-10-01 proposed_class: contractual fact scope: client-north status: needs-review reason: conversation references a signed agreement not present in export review_owner: account director promotion_blocker: obtain agreement and verify section ``` The model may extract candidates. It must not approve its own extraction as business truth. Promotion requires the named source and owner. ## Deduplicate without erasing provenance The same decision may appear in a chat, a Notion summary, an email copy, and an agency report. Do not keep four competing current records. Choose one canonical record and list the other items as evidence or superseded derivatives. Use exact hashes for identical files. Use titles, dates, source IDs, and human review for near-duplicates. Similar language does not prove two records have the same authority or scope. ```yaml # owned/decisions/DEC-2026-081.yaml record_id: DEC-2026-081 statement: Client North renews on 2026-10-01 status: accepted authority: contractual canonical_source: contracts/client-north-signed.pdf#section-4.2 supporting_evidence: - chatgpt:conv-4481#message-19 - notion:page-182#renewal-note supersedes: - agency:file-771 owner: account director scope: client-north ``` Supersession preserves history. Deletion is a separate action controlled by retention, legal, security, and business rules. ## Reconcile export coverage and missing material Treat coverage as a testable claim. Make an expected-source list from invoices, account rosters, workspace membership, project lists, old links, browser bookmarks, and the people who performed the work. Compare that list with the exports received. For each source, record the earliest and latest item dates, record count, attachment count, database or page count where available, and known exclusions. Ask whether shared projects, deleted items, archived spaces, custom instructions, saved memories, uploaded files, comments, and automation configuration are included. Provider documentation can answer some questions. Only the actual package can show what arrived for this account. Do not let an absent record become an inferred record. If the archive references a signed agreement that is missing, create a coverage gap with an owner and recovery action. If an agency says a workflow lived in a private tool that it cannot export, record the dependency and decide whether screenshots, a walkthrough, configuration copy, API extraction, or a clean rebuild can satisfy acceptance. Use explicit states: `confirmed-present`, `expected-missing`, `not-applicable`, `provider-excluded`, `access-blocked`, and `unknown`. A migration can proceed around a noncritical gap, but the handoff receipt must preserve the gap and its effect on the survival job. ## Protect sensitive data during the move Exports can concentrate years of customer data, private conversations, files, and credentials in one package. Limit who can request, download, extract, and review them. Use approved storage with access logging and encryption appropriate to the business. Do not place raw packages in a broadly shared repository. Scan for secrets before moving converted content into version control or an agent-readable folder. A transcript may contain an API key, reset link, personal address, health detail, or payment information that was never intended as durable context. Quarantine suspected secrets and sensitive records for an authorized owner. Preserve enough provenance to make a decision without copying sensitive content into every derivative. An inventory can use an item ID, classification, owner, and restricted path. A receipt can record a file hash and result. Neither needs to reproduce the private source text. Set a retention decision for temporary extraction directories and review copies. Keep the immutable raw package only where the business has decided it belongs. Delete temporary copies only after the canonical destination, backup, and proof are accepted. Deletion itself needs a receipt when the material is sensitive or the migration contract requires proof. ## Use a parallel-run window for important jobs For a high-value recurring job, run the owned path alongside the current vendor path before cutover. Use the same approved input and compare facts, required sections, source citations, stopping behavior, and action count. Do not require matching prose. Investigate every material difference. The old system may contain an undocumented rule. The new system may be missing a current source. The old result may be wrong. Parallel output is evidence for review, not proof that the vendor version wins. Set the window in advance. Two or three cycles may be enough for a monthly report only if fixtures cover failure cases between live runs. A daily workflow can provide more natural repetitions. Define who reviews differences, what blocks cutover, and what evidence closes each gap. Keep both paths draft-only during comparison when an external action would duplicate work. If the old system must remain the live executor, prevent the new path from holding the same send or write capability. The test should compare preparation while one controlled path owns execution. End the parallel window with an acceptance decision. Record the last accepted old-system run, first accepted owned-system run, unresolved exceptions, rollback window, and owner. Avoid an indefinite dual system where corrections continue to split across both places. ## Rebuild one recurring job The migration becomes operational when one job runs from owned records without the vendor archive. Write a task brief that names the result, reader, allowed sources, source order, output, acceptance checks, approval owner, and stop. Then prepare a small context packet from promoted records. Do not point the new agent at the entire converted archive. ```yaml # owned/jobs/monthly-client-report/brief.yaml job: prepare monthly Client North report draft allowed_sources: - owned/clients/client-north/current.md - owned/decisions/ - owned/procedures/monthly-report.md - owned/corrections/accepted.md source_order: - accepted contractual and business decisions - current client facts - accepted procedure - accepted corrections archive_role: evidence only output: owned/drafts/client-north/monthly-report.md acceptance: - every status claim cites an owned source - current renewal date is correct - no other client data appears - no external send occurs stopping_point: draft and receipt written ``` The archive may remain searchable for investigation. It should not enter every routine run. This smaller packet reduces stale-source collisions and makes missing records visible. ## Carry corrections and open work separately Agency archives often hide corrections inside comments and later chat turns. Extract the wrong behavior, accepted replacement, scope, owner, effective date, and test. Keep proposed corrections separate from accepted corrections. The article on [how corrections become system memory](/articles/corrections-become-system-memory) makes the next run the proof. A correction is not migrated until the fixture that previously failed now passes. Open work needs its own register. Do not promote a tentative plan into a procedure because it was the last message in a thread. Record the owner, next action, blocker, source, and review date. The new operator can then distinguish current commitments from ideas. ## Build the acceptance test before cutting access Run the survival job from a clean session with no old conversation and no vendor search access. ```text # owned/jobs/monthly-client-report/acceptance.fixture SETUP clean directory and fresh model session DENY access to ChatGPT, Claude, Notion, and agency archive GIVE only brief.yaml and allowed owned sources EXPECT current renewal date from DEC-2026-081 EXPECT current reporting procedure EXPECT accepted correction COR-2026-014 EXPECT zero cross-client canary strings EXPECT draft at defined path EXPECT no send or CRM write EXPECT receipt with source paths, versions, tests, and stop ``` Run the test with the incoming operator. A migration performed only by the outgoing agency does not prove handoff. The buyer or replacement worker should locate the runbook, execute the fixture, explain any blocked state, and identify where corrections enter. The [AI system ownership transfer checklist](/articles/ai-system-ownership-transfer-checklist) expands this into controlled change, rollback, credential rotation, and independent operation. [Your AI system is not yours until it can transfer](/articles/transfer-the-repo-client-handoff) distinguishes file delivery from a working handoff. ## Separate model replacement from record migration The business records should remain stable while provider adapters change. Put message roles, tool schemas, model identifiers, API parsing, and provider authentication in adapter files. Keep decisions, procedures, correction rules, and acceptance fixtures outside them. Run the same fixture with the replacement model. The prose may differ. The facts, boundaries, required sections, and stop should remain consistent. The guide to [switching models without losing business context](/articles/switch-ai-models-without-losing-context) covers the cross-model fixture. [Replacing the model while keeping business memory](/articles/replace-the-model-keep-business-memory) shows the stable-record and variable-adapter split. Do not cancel a subscription merely because the export exists. The [AI subscription shutdown test](/articles/ai-subscription-shutdown-test) inventories account dependencies, moves durable records, tests the replacement, and distinguishes cancellation from deletion. ## Document credentials without copying secrets The handoff needs an account and credential map. It should name each system, owner, role, recovery path, storage location, rotation requirement, and deprovisioning owner. Do not place API keys or passwords in the archive or runbook. ```yaml # owned/handoff/access-map.yaml systems: - system: reporting repository owner: operations director required_role: maintainer secret_location: company password manager/reporting-repo rotation_after_handoff: true - system: email sender owner: sales owner required_role: none for draft-only job secret_location: not provisioned rotation_after_handoff: false ``` Transfer access through the business's approved identity and secret systems. Test that the incoming operator can perform the accepted job with least privilege. Remove outgoing access only after the owner authorizes the exact account changes and recovery paths are proven. ## Rehearse the handoff with a person who did not build it Give the incoming operator the same materials a future replacement would receive: root path, inventory, record map, runbook, task brief, fixtures, access map, and support boundary. Do not supplement the rehearsal with the outgoing builder's remembered shortcuts. Ask the operator to locate the canonical record for one important claim, explain which source wins, run the survival fixture, diagnose a staged missing-source failure, apply a proposed correction through review, and identify where a consequential action would require approval. Observe where the operator guesses. A path that makes sense only to its creator needs a clearer index. A test that requires an undocumented environment variable needs setup instructions. A correction path that ends in a private vendor queue has not transferred. Revise the owned files and repeat the failed step. The rehearsal ends with two receipts. The operator records what they read, ran, changed, and could not complete. The reviewer records acceptance by item. Store open support needs as bounded work, not as a promise that the outgoing vendor will remain available indefinitely. ## Verify the system after the exit After authorized account changes occur, run the survival fixture again. Verify that the owned job no longer calls the old provider, reads the old workspace, depends on the outgoing agency's identity, or writes receipts into an inaccessible location. Inspect scheduled jobs, webhooks, API keys, browser extensions, shared links, email forwarding, and automation accounts that might retain the old dependency. Search current configuration and runtime logs. Documentation that says "migrated" does not prove a scheduled process changed. Run a backup restore after cutover. Confirm that the latest accepted records, runbook, fixtures, and receipts restore together. Check that the incoming owner controls repository administration, storage billing, domain or account recovery, and the credential manager entries needed for the job. Set a short post-exit review date. Examine blocked runs, correction recurrence, missing sources, and unexpected calls to retired systems. Close the migration only when the owner accepts remaining exceptions and knows who maintains the new path. ## Record the final handoff The final receipt should list raw export hashes, inventory counts, conversion errors, promoted records, rejected or pending candidates, survival-job result, restore result, incoming operator, open risks, and actions still awaiting approval. The [build acceptance list](/articles/what-a-build-acceptance-list-contains) gives a concrete format for files, tests, receipts, permissions, failure cases, and handoff proof. Acceptance should be binary for each item. "Mostly migrated" hides which business dependency remains in the old system. Keep the raw export, readable derivative, and maintained record as three distinct layers. The raw export supports audit and reprocessing. The readable derivative supports review. The maintained record supports current work. ## Approval and stopping boundary Migration work may request approved exports, copy files, hash packages, inventory records, create readable derivatives, extract review candidates, prepare owned records, run offline fixtures, and write receipts. It stops before deleting provider or agency data, closing accounts, revoking users, rotating credentials, changing retention periods, widening access, sending customer communications, publishing, deploying, or replacing a current system. Those actions require a named target, owner, recovery path, timing, and fresh approval. The process also stops when export coverage is unknown, a sensitive record lacks an approved destination, source authority conflicts, the incoming operator cannot pass the fixture, or the business lacks access to a required account. The receipt names the unresolved dependency. ## Where to start Start with the smallest exit that proves something. Choose one recurring job, preserve the exports it depends on, and run the acceptance fixture from owned records. Prove that one valuable job survives without the old account before you fund a larger archive migration. ## Sources - [OpenAI: Export ChatGPT history and data](https://help.openai.com/en/articles/7260999-how-do-i-export-my-chatgpt-history-and-data) - [Anthropic: Export Claude data](https://support.claude.com/en/articles/9450526-export-your-claude-data) - [Notion: Export your content](https://www.notion.com/help/export-your-content) - [Git: Getting a Git Repository](https://git-scm.com/book/en/v2/Git-Basics-Getting-a-Git-Repository) ## Questions about The Complete Guide to Leaving Vendor AI Memory ### How do I leave an AI service without losing years of context? Preserve the provider's raw export first, then inventory chats, files, saved memory, instructions, projects, and connected records. Promote only current sourced facts, decisions, procedures, corrections, and open work into maintained files before rebuilding one recurring job. ### Should I delete my old AI account as soon as the export arrives? Keep the old system available during a bounded parallel-run window when the job matters. Validate export coverage, rebuild the recurring job, test the replacement, document missing material, then handle cancellation and deletion as separate verified steps. ### Can I just upload my entire chat archive into the new AI? Do not promote the raw archive directly into governing memory because it mixes drafts, stale instructions, unsupported claims, and conflicting decisions. Preserve it as evidence, then classify extracted records as keep, reject, or review with provenance intact. ## Save the visual summary Define the exit before exporting | Preserve the raw exports | Build an inventory before reading everything | Convert into a readable archive | Classify before promoting [Download the PNG](/infographics/leaving-vendor-ai-memory-guide-1200x1500.png) ## Related: Ownership and Migration - [How to Turn ChatGPT History Into Business Memory](/articles/migrate-chatgpt-history-to-business-memory) - [How to Move Years of AI Work Out of a Vendor Archive](/articles/move-ai-work-out-of-vendor-archive) - [How to Move ChatGPT Memory to Claude Without Importing Old Mistakes](/articles/move-chatgpt-memory-to-claude) ---- ```text .|########||. .|########||. .|########||. |##||. .||##|. |##||. .||##|. |##||. .||##|. |#|. .|#|. |#|. .|#|. |#|. .|#|. |#| |#| |#| |#| |#| |#| .#| |#. .#| |#. .#| |#. .#. .#| .#. .#| .#. .#| |#. .#| |#. .#| |#. .#| |# ...... #| |# ...... #| |# ...... #| .# ||#########| #| .# ||#########| #| .# ||#########| #| .#. |######||######|. .#. .#. |######||######|. .#. .#. |######||######|. .#. #. .##|###|##|#|######| .# #. .##|###|##|#|######| .# #. .##|###|##|#|######| .# || |##|#||||||||||||#||#| || || |##|#||||||||||||#||#| || || |##|#||||||||||||#||#| || # |#||||||||||||||||||||#| #. # |#||||||||||||||||||||#| #. # |#||||||||||||||||||||#| #. || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || # .#||||||||||||||||||||||||#| # # .#||||||||||||||||||||||||#| # # .#||||||||||||||||||||||||#| # || ....|||#||||||##|#|||#|#||##||||....|. || || ....|||#||||||##|#|||#|#||##||||....|. || || ....|||#||||||##|#|||#|#||##||||....|. || #. . ....|# ....#||| ||| .#||| ....#. .# #. . ....|# ....#||| ||| .#||| ....#. .# #. . ....|# ....#||| ||| .#||| ....#. .# # . ||||#| .#####|| . .#| .#||| ||||# #. # . ||||#| .#####|| . .#| .#||| ||||# #. # . ||||#| .#####|| . .#| .#||| ||||# #. .| ||||| #. |#|||||. || #. |#||. .||||| || .| ||||| #. |#|||||. || #. |#||. .||||| || .| ||||| #. |#|||||. || #. |#||. .||||| || || |.... #. ....|#. |. ...|. ....|| || || |.... #. ....|#. |. ...|. ....|| || || |.... #. ....|#. |. ...|. ....|| || #. .||||||##||||||##||####|||||||#||||||#| .# #. .||||||##||||||##||####|||||||#||||||#| .# #. .||||||##||||||##||####|||||||#||||||#| .# # .||####|###########################|. # # .||####|###########################|. # # .||####|###########################|. # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#||||||############|######|||||#| # # .#||||||############|######|||||#| # # .#||||||############|######|||||#| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# || | ...|| ...##| | #| .|. |. ##### .. .| || || | ...|| ...##| | #| .|. |. ##### .. .| || || | ...|| ...##| | #| .|. |. ##### .. .| || || ||||| || ||||#| . |. |. |#. ||||| |#| || || || ||||| || ||||#| . |. |. |#. ||||| |#| || || || ||||| || ||||#| . |. |. |#. ||||| |#| || || .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. #.|||||||#############################| ||| .# #.|||||||#############################| ||| .# #.|||||||#############################| ||| .# || ##||#||#||||||#||#||#||#||##. || || ##||#||#||||||#||#||#||#||##. || || ##||#||#||||||#||#||#||#||##. || # .#|||||||||#||#|||||||||||#| # # .#|||||||||#||#|||||||||||#| # # .#|||||||||#||#|||||||||||#| # || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || .# |#||||||||||||||||||||#| #. .# |#||||||||||||||||||||#| #. .# |#||||||||||||||||||||#| #. || |#||||||||||||||||||#| || || |#||||||||||||||||||#| || || |#||||||||||||||||||#| || #. .######|#|#########| .# #. .######|#|#########| .# #. .######|#|#########| .# .#. |#####||||#####| .#. .#. |#####||||#####| .#. .#. |#####||||#####| .#. |# ||########|| #. |# ||########|| #. |# ||########|| #. |# ...... #| |# ...... #| |# ...... #| |#. .#| |#. .#| |#. .#| |#. .#. |#. .#. |#. .#. .#| |#. .#| |#. .#| |#. |#| |#| |#| |#| |#| |#| .|#|. .|#|. .|#|. .|#|. .|#|. .|#|. |##||. .||##| |##||. .||##| |##||. .||##| .||########||. .||########||. .||########||. Scale With Search 2026 [scalewithsearch.com](https://scalewithsearch.com) ```