Comparison

Kelu vs Kapa.ai

Both platforms turn your documentation into an AI-powered assistant for developers. This comparison covers retrieval architecture, integration breadth, SDK access, and pricing model — based on publicly available information.

Feature comparison

Information about Kapa.ai is based on their public website and documentation.

Feature
Kelu
Kapa.ai
Retrieval method
Hybrid: pgvector semantic + BM25 full-text, RRF fusion, LLM query rewrite, cross-encoder re-ranking
Semantic vector search
Website / docs crawling
Yes — sitemap and link discovery, incremental sync
Yes
GitHub repo indexing
Yes — READMEs, docs/, ADRs, changelogs, plus issues, pull requests and discussions. Git-push webhook for incremental sync.
Yes
Slack bot
Yes — /askai slash command and @mention. Included on Pro and Enterprise.
Yes
Discord bot
Yes — slash command and @mention. Included on Pro and Enterprise.
Yes
Web widget
Yes — hosted at widget.kelu.dev. reCAPTCHA bot protection. Full CSS override.
Yes
React component
Yes — @kelu/react. <KeluChat /> and useKelu headless hook.
Available
JavaScript / Node SDK
Yes — @kelu/sdk. Framework-agnostic. Targets web, React Native, Node.js.
Available
Open REST API
Yes — fully documented. WebSocket streaming. No gating on API access.
Available on higher tiers
Collections & per-collection permissions
Yes — group sources, scope client keys per collection
Project-level scoping
Versioned documentation
Yes — v1, v2, latest tags on sources. Per-version client keys.
Not documented
Suggested questions & related articles
Yes — included in every chat response
Varies by plan
Analytics & daily trends
Yes — conversations, feedback ratio, gaps, daily trends on all plans
Yes — analytics dashboard
AI evals (groundedness, citation quality)
Yes — automated LLM-as-judge evals, scores exportable via API
Feedback tracking
Pricing transparency
Public pricing page. Free plan included. No per-seat fees.
Contact for pricing

Where Kelu focuses differently

Hybrid retrieval with LLM query intelligence

Kelu combines pgvector cosine similarity, BM25 full-text, and reciprocal rank fusion. Before retrieval, an LLM rewrites and expands the user query to improve recall. A cross-encoder re-ranker then sorts the top candidates by relevance to the rewritten query.

Full API and SDK openness

Every feature available in the widget is accessible via the REST API and SDKs with no tier gating on API access. This means you can build custom mobile experiences, server-side integrations, and automated pipelines from the start.

Collections and versioned docs

When your product has multiple versions or your team serves different audiences, collections let you partition knowledge and issue scoped client keys. Users querying v1 docs never receive answers from v2 content.

Transparent public pricing

Kelu publishes its pricing openly. There is no requirement to contact sales before you can evaluate the product cost. A free tier is included without a credit card.

Frequently asked questions

How is Kelu different from Kapa.ai?

Both platforms answer questions from your documentation with citation-backed AI. Kelu differentiates on retrieval (hybrid pgvector + full-text with RRF fusion, LLM query rewriting, and re-ranking), ungated API/SDK access on every plan, collections with per-key permissions, versioned docs, and public pricing with a free tier.

Can I migrate from Kapa.ai to Kelu?

Yes. Kelu indexes the same kinds of sources — docs sites, GitHub repos, Notion, Confluence, and resolved helpdesk tickets — so migration is usually just reconnecting your sources and swapping the widget embed or SDK calls. Most teams are running in under an hour.

Does Kelu support the same integrations?

Kelu ships a web widget, React component, JavaScript/Node SDK, Slack and Discord bots, an open REST API with WebSocket streaming, an MCP server per knowledge base, and helpdesk integrations for ticket deflection and copilot drafts.

Do answers include citations?

Yes. Every answer links to the exact source passages it was grounded in, and automated LLM-as-judge evals continuously score groundedness and citation quality.

Is my data used to train AI models?

No. Your connected content and your users' conversations are never used to train models. Content is indexed solely to answer questions for your knowledge base, and PII masking and retention controls are built in.

How much does Kelu cost?

Pricing is public — there is a free plan with no credit card required, and paid plans scale with usage rather than per-seat fees. See the pricing page for current details.

Is this comparison up to date?

The Kapa.ai column is based on publicly available information from their website and documentation. If you spot something outdated, contact us and we'll correct it.

See for yourself

Connect your docs and get your first AI-powered answer in under 5 minutes. No credit card required.