> ## Documentation Index > Fetch the complete documentation index at: https://sitegpt.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. # Playbook: Knowledge maintenance > An agent playbook for auditing, updating, resyncing, and cleaning chatbot knowledge with the SiteGPT CLI. Use this playbook for recurring knowledge-base maintenance, failed-ingestion cleanup, source updates, and content quality checks. ## Required access Recommended scopes: ```bash theme={null} sitegpt login \ --profile knowledge-agent \ --scope account:read \ --scope chatbots:read \ --scope knowledge:read \ --scope knowledge:write \ --scope knowledge:delete ``` Add `settings:read` if the agent should inspect chatbot settings and `settings:write` if it should update crawler-related settings. ## 1. Inspect health ```bash theme={null} sitegpt dashboard --chatbot --json sitegpt knowledge documents stats --chatbot --json sitegpt knowledge documents list --chatbot --status FAILED --json sitegpt knowledge documents list --chatbot --status PROCESSING --json sitegpt knowledge sync-jobs list --chatbot --json ``` Look for: * Failed documents * Stale source documents * Missing sitemap sync/scan jobs * Duplicated documents * Outdated custom responses * Large pending queues ## 2. Resync failed or stale documents Resync all failed documents: ```bash theme={null} sitegpt knowledge documents resync --chatbot --state failed ``` Resync by source: ```bash theme={null} sitegpt knowledge documents resync --chatbot --source WEBSITE sitegpt knowledge documents resync --chatbot --source SITEMAP ``` Resync exact documents: ```bash theme={null} sitegpt knowledge documents resync --chatbot --document --document ``` ## 3. Update document config For pages with layout noise: ```bash theme={null} sitegpt knowledge documents update-config \ --chatbot \ --document \ --only-main-content true \ --include-selector main \ --exclude-selector nav \ --exclude-selector footer ``` For recurring content refresh: ```bash theme={null} sitegpt knowledge documents update-config --chatbot --document --sync WEEKLY ``` Valid sync frequencies: ```text theme={null} NEVER, DAILY, WEEKLY, MONTHLY ``` ## 4. Maintain sync jobs List active jobs: ```bash theme={null} sitegpt knowledge sync-jobs list --chatbot ``` Update job frequency: ```bash theme={null} sitegpt knowledge sync-jobs update --chatbot

--sync WEEKLY sitegpt knowledge sync-jobs update --chatbot

--scan WEEKLY ``` Disable a job: ```bash theme={null} sitegpt knowledge sync-jobs delete --chatbot

--yes ``` Sitemap `--scan` keeps sitemap URL additions/deletions up to date. `--sync` refreshes content for known URLs. ## 5. Add new website content Prefer sitemap: ```bash theme={null} sitegpt knowledge sitemap add \ --chatbot \ https://example.com/sitemap.xml \ --max-links 500 \ --only-main-content true \ --sync WEEKLY \ --scan WEEKLY ``` Fallback to website crawl: ```bash theme={null} sitegpt knowledge website add \ --chatbot \ https://example.com/docs \ --depth 3 \ --max-links 200 \ --only-main-content true ``` Use exact links for surgical updates: ```bash theme={null} sitegpt knowledge links add \ --chatbot \ https://example.com/pricing \ https://example.com/refund-policy \ --only-main-content true ``` ## 6. Maintain source connections List sources: ```bash theme={null} sitegpt knowledge sources list --chatbot --json ``` Select more files for an existing picker connection: ```bash theme={null} sitegpt knowledge sources authorize --chatbot

sitegpt knowledge sources documents --chatbot

sitegpt knowledge sources ingest --chatbot

``` Ingest selected documents only: ```bash theme={null} sitegpt knowledge sources ingest \ --chatbot \ \ --external-id \ --sync WEEKLY ``` GitHub: ```bash theme={null} sitegpt knowledge sources github repos --chatbot --source sitegpt knowledge sources github files --chatbot --source --owner --repo sitegpt knowledge sources ingest --chatbot

--repo --branch main --pattern "docs/**" ``` Confluence: ```bash theme={null} sitegpt knowledge sources confluence spaces --chatbot --source sitegpt knowledge sources confluence pages --chatbot --source --space sitegpt knowledge sources ingest --chatbot

--page ``` ## 7. Add or update text knowledge Use text knowledge for short curated policy text, summaries, or controlled notes: ```bash theme={null} sitegpt knowledge text update --chatbot --name "Support policy notes" --file ./support-notes.md ``` Because this replaces the text document, read existing content first when editing: ```bash theme={null} sitegpt knowledge documents list --chatbot --source TEXT --json sitegpt knowledge documents get --chatbot

--content --json ``` ## 8. Maintain Custom Responses List open Custom Responses: ```bash theme={null} sitegpt knowledge custom-responses list --chatbot --state OPEN --json ``` Add a new approved answer: ```bash theme={null} sitegpt knowledge custom-responses add \ --chatbot \ --question "Do you offer refunds?" \ --answer "Refund eligibility depends on your plan and billing terms. Contact support@example.com for account-specific help." ``` Update an existing response: ```bash theme={null} sitegpt knowledge custom-responses update --chatbot

--answer "Updated approved answer." ``` Delete outdated responses: ```bash theme={null} sitegpt knowledge custom-responses delete --chatbot

--yes ``` ## 9. Delete obsolete documents carefully Preview matching documents: ```bash theme={null} sitegpt knowledge documents list --chatbot --query "old product" --json ``` Delete exact IDs: ```bash theme={null} sitegpt knowledge documents delete --chatbot --document --document --yes ``` Delete failed documents only: ```bash theme={null} sitegpt knowledge documents delete --chatbot --state failed --yes ``` Avoid `--all --yes` unless the user explicitly asked for broad deletion. ## 10. Final verification ```bash theme={null} sitegpt dashboard --chatbot --json sitegpt knowledge documents stats --chatbot --json sitegpt knowledge documents list --chatbot --status FAILED --json sitegpt messages send --chatbot "What topics are you trained on?" --json ``` Report: * What was added or updated * Any failed documents * Any sync jobs changed * Any deleted documents or custom responses * Follow-up work requiring human approval