docs: reconstruct bilingual guides and refresh high-resolution screenshots

[yangchuansheng]

Summary

The documentation mixes outdated navigation, workflows, and screenshots, making it difficult to follow current FastGPT tasks. Reorganize the bilingual documentation around reader tasks and update the tutorials with verified SaaS workflows and native high-resolution screenshots.

• Reconstruct navigation and discovery across 285 Chinese/English page pairs, add missing task guides, and retain historical-version context. Preserve valid topic URLs and existing anchors, with direct permanent redirects for consolidated entries.
• Refresh getting started, Agents, Skills, knowledge-base tags, publishing, permissions, workflow nodes, chat rendering, evaluation, and system-tool debugging. Add reproducible tutorial fixtures and fix their static download routing.
• Complete the knowledge-base task chain across 12 bilingual page pairs: import material, test retrieval, connect an Application, publish, update content, and recover from an empty search. Add four focused how-to guides and clarify retrieval parameters, template handling, and the distinction between content indexing and configuration publication.
• Publish 135 current high-resolution images from original capture assets or native ego-browser screenshots, with source and SHA-256 records. Improve proportional article-image sizing and retain click-to-zoom.
• Add focused documentation, navigation, language, URL, sitemap, and tutorial-download checks; address localized navigation and 404 behavior.

Latest documentation rewrite

Commit 3f0408a96 updates 92 official documentation files across getting started, core concepts, knowledge-base lifecycle, Agent/workflow orchestration, self-hosting, and API integration.

• Verify the bilingual SSE event tables against all 17 runtime event types and correct interactive continuation to reuse chatId and submit user input as a message.
• Correct member-authenticated collection access, signed-link handling, token-based citation limits, reference-prompt configuration, and source-sync/reindexing lifecycle descriptions.
• Align sandbox lifecycle and workflow edge states with the implementation, replace unsupported capacity guarantees with workload-based validation, and clarify backup and upgrade procedures.
• Restore 345 historical heading-anchor entry points across the edited pages and replace suppressed SDK type errors with a typed request example.
• Keep engineering notes under docs/, local indexes, and environment configuration outside this commit.

Validation of the latest commit

• pnpm --dir document check:docs: passed for 570 MDX files, 285 English pages, 44 navigation files, and 651 English links.
• pnpm --dir document test:checks: all 12 tests passed.
• pnpm --dir document typecheck: passed.
• pnpm --dir document build: passed, generating 579 static pages.
• pnpm --dir document test:site: all 12 HTTP acceptance tests passed against the local production build.
• Both TypeScript SDK snippets passed type checking against the installed OpenAI SDK 4.104.0; Python snippets passed syntax checks. Both SSE tables match the runtime event enum.
• Historical-anchor comparison, added-line credential-pattern screening, and git diff --check passed.

Earlier validation evidence

The following checks belong to earlier iterations and retain their original scenario scope:

• pnpm --dir document check:docs passed before the latest rewrite: 570 MDX files, 285 English pages, 44 navigation files, and 712 English links.
• Type validation and the latest production build passed; the build generated 579 static pages.
• All 12 documentation-check tests and all 12 HTTP site-acceptance tests passed for the knowledge-base implementation. The subsequent SaaS procedure corrections passed documentation checks, a fresh production build, and browser checks of all 10 changed localized routes.
• All 24 knowledge-base article routes were inspected in ego-browser. The batch preserves all 203 existing heading IDs; desktop tables, media, contextual links, and mobile task navigation were checked.
• Four key scenarios passed on official SaaS in user-authorized isolated test resources: a published 15-day travel-policy answer with supporting citations; explicit handling of a missing hotel-allowance standard; a chunk/index update from 15 to 30 days verified in retrieval and a new shared conversation; and disabling/re-enabling the test Collection, with empty-search and restored-answer checks.
• SaaS acceptance used 阿里-emb4 and qwen3.8-flash, semantic search, a citation budget of 5000, minimum relevance 0, and reranking/question optimization off. Execution followed the Chinese procedures; English counterparts received semantic and rendered reading-path review.
• Earlier desktop and mobile browser checks covered the refreshed bilingual pages, image loading and aspect ratios, overflow, and image zoom/Escape behavior. System-tool changes passed in both languages at 1440×900 and 390×844.
• Standards and Spec reviews passed for each delivered batch, including separate follow-up reviews of the SaaS corrections. Those incremental diffs passed git diff --check.

Remaining scope and verification limits

The latest rewrite received source review, static checks, and local production-site acceptance. Live model calls, production deployment procedures, and a fresh end-to-end SaaS walkthrough remain pending for this commit. The build retains the existing Next.js multiple-lockfile workspace-root warning. Earlier screenshots and SaaS results below remain historical evidence for the scenarios tested at that time.

The complete screenshot refresh remains open. The remaining inventory contains 213 resources: 9 diagrams/icons, 36 historical-version resources, and 168 resources awaiting matching originals or environments. This inventory includes existing high-resolution images and is not a count of low-resolution screenshots. Remaining sources include self-hosted administration/SSO, Sealos, AI Proxy, third-party knowledge-base backends, invitation acceptance, domain status, and historical versions.

The knowledge-base runtime acceptance covers the local TXT fixture and the four scenarios above. CSV/XLSX import, replacement uploads, reprocessing, supplementary-index changes, reranking failures, alternate search modes, Workflow filters, other publishing channels, independent logged-out-session behavior, and self-hosted environments retain their recorded source-review or verification limits. Model wording can vary; acceptance checks facts and cited passages.

The new system-tool dialog shows the state before generating a debug link. Debug-link generation, CLI connectivity, third-party callbacks, deployment workflows, and the full permission matrix retain their documented validation limits. The full repository test suite awaits final acceptance. Detailed evidence and source records are maintained under .agents/design/documentation/.

[github-actions]

✅ Docs preview deployed

🔗 👀 Click here to visit preview

ghcr.io/labring/fastgpt-docs-pr:3f0408a96dc175b9216641cd38d14f85606a34f6

🕒 Time: 2026-10-19 00:26:03 (UTC+8)