Knowledge Base
The knowledge base is the long-term memory of NestFleet. It stores articles, runbooks, FAQs, known issues, and resolved case summaries. Every auto-reply and triage decision is grounded in this knowledge — keeping it accurate and up to date directly improves AI quality.
What belongs in the knowledge base
The knowledge base accepts several types of content:
| Type | Description |
|---|---|
| Article | A free-form document: FAQ entry, how-to guide, configuration reference, or troubleshooting runbook. Supports Markdown formatting. |
| Known Issue | A documented bug or limitation with a known workaround or fix. Linked to one or more cases and optionally to a change request. |
| Resolved Case (auto) | A summary of a resolved case captured automatically. Includes the original issue, the resolution steps, and the final reply sent to the customer. |
| GitHub File (sync) | A file from the product's GitHub repository (e.g. README, CHANGELOG, runbook) synced periodically and indexed for search. |
How the knowledge base feeds triage
All knowledge base content is embedded using the configured embedding model and stored as vectors in PostgreSQL (via the pgvector extension). When a new case arrives, the triage and auto-reply agents perform a vector similarity search to find the most relevant articles.
This retrieval-augmented generation (RAG) approach keeps AI responses grounded in your actual product documentation rather than the model's general training data. The quality of the search results — and therefore the auto-replies — depends directly on:
- The accuracy and completeness of your articles
- The granularity of known issues (one issue per article, not dozens per article)
- The freshness of resolved case summaries
- The quality of the embedding model (see Environment Variables for embedding config)
A newly deployed NestFleet instance has an empty knowledge base. Auto-reply will not trigger for cases until at least some articles or known issues exist. Start by importing your existing FAQ, adding your most common support scenarios, and enabling resolved-case capture.
Creating articles manually
Navigate to Knowledge Base in the sidebar, then click New Article. Each article has:
- Title — used in search results and in AI-generated reply citations
- Content — Markdown body. Be specific and include the exact error messages and workaround steps that a customer would need.
- Tags — optional labels for filtering and grouping (e.g. "authentication", "billing", "api")
- Visibility — Internal only (team-facing) or Public (can be linked in replies to customers)
- Product scope — articles are scoped to a specific product or shared across all products
Once saved, the article is re-embedded in the background (a pg-boss job). It becomes searchable within 30–60 seconds on most configurations.
Known issues
A Known Issue is a special article type that represents a confirmed bug or limitation. It has additional fields compared to a standard article:
- Status — open, in-progress, resolved, won't-fix
- Workaround — if a workaround exists, describe it here; this is injected into auto-replies
- Fix version — once resolved, the version in which it was fixed
- Linked cases — all cases that have been matched to this known issue
- Linked change request — the CR tracking the fix, if one exists
To link a case to a known issue, open the case detail, click Link Known Issue, and search for the relevant issue. Once linked, future cases that match the same known issue will automatically inherit the workaround text in their auto-reply.
Auto-update proposals
When a case is closed with a resolution note, NestFleet compares the resolution to existing knowledge base articles and known issues. If the resolution contains new information not captured in any existing article, the system creates a Knowledge Update Proposal — a suggested edit to an existing article or a draft for a new article.
Proposals appear in the Knowledge Base → Pending Updates view. A team member with Knowledge Lead access reviews the proposal and can:
- Accept — apply the proposed changes to the article or create the new article
- Edit and accept — revise the proposal before applying
- Reject — discard the proposal (e.g. if the resolution was one-off or already documented)
This feedback loop means the knowledge base grows richer with every case resolved, without requiring manual curation effort for routine cases.
Importing from GitHub
If your product has a GitHub repository connected, you can configure NestFleet to periodically sync specific files or directories into the knowledge base. Go to Settings → Knowledge Sources → GitHub Sync and specify the file paths to import (e.g. docs/**/*.md,CHANGELOG.md, README.md).
Synced files are re-embedded every 24 hours (or on webhook push if the GitHub App is configured). Changes to the source files in the repository are reflected in the knowledge base automatically.