Now with agentic verification — answers you can trust

Your knowledge base, answered instantly.

dysha.ai connects to your docs, repos, tickets, and wikis — then answers technical questions with verified citations. No hallucinations. No guessing. Just answers, backed by your actual knowledge.

Secure cloud · Your data is never used for training · Always cited

app.dysha.ai

Sources connected

0

GitHub · Notion · Jira

Cited answers

0%

every response verified

Avg response

1.4s

sub-2s latency

How does our auth middleware handle token refresh?
Searching knowledge baseauth-service · 3 sources

The middleware uses a sliding window refresh strategy — tokens within 5 min of expiry are silently renewed via response headers.

Sourcesauth/middleware.ts:142docs/auth-design.md

50+

Knowledge connectors

< 2s

Median response time

100%

Answers with citations

< 30m

Time to first deployment

Connects to your existing stack

GitHubConfluenceJiraSlackGitLabNotionZendeskLinearIntercomGoogle Drive

How it works

From question to verified answer in seconds

dysha.ai doesn't guess. It searches your actual knowledge base, evaluates the evidence, and only answers when confident.

01

Connect your knowledge sources

Point dysha.ai at your GitHub repos, Confluence spaces, Jira projects, Slack channels, or any of our 50+ connectors. We index everything automatically and keep it in sync.

GitHub · Confluence · Jira · Slack · GitLab · Notion · Zendesk · OpenAPI specs · File uploads

02

dysha.ai retrieves and verifies

Every question triggers an intelligent multi-step retrieval pipeline: search the knowledge base, evaluate the evidence, retry with a refined query if needed, then synthesise an answer — all before writing a single word.

Hybrid search · Semantic reranking · Evidence validation · Knowledge-base grounding

03

Deliver cited answers everywhere

Answers arrive with exact source citations — file path, page URL, or issue link. Deploy as a website widget, Slack bot, MCP server, or via API. Your team gets answers where they already work.

Web chat · Slack bot · Embeddable widget · MCP server · REST API

Features

Everything you need. Nothing you don't.

Built for engineering teams who need accurate answers, not impressive demos.

Intelligent retrieval

dysha.ai runs a multi-step search pipeline — search, evaluate, and refine — before synthesising an answer. Inconclusive first pass? It retries with a sharper query automatically.

Cited answers only

Every answer includes exact source citations: file path, line number, page URL, or issue link. If there's nothing in the KB, it says so — it never invents an answer.

True multi-tenancy

Every chunk is stamped with a tenant ID. Cross-org data leakage is architecturally impossible — isolation is enforced at the vector store level, not application logic.

Self-hosted or cloud

Run dysha.ai entirely on your own infrastructure. Your vectors, your LLM API key, your data. Or use dysha.ai Cloud if you prefer zero-ops. You choose.

Hybrid retrieval

Dense vector search combined with keyword search, fused and reranked for best-in-class accuracy on technical content. Fast and precise on every query.

Gap analytics

See exactly which questions your KB can't answer. Coverage gap analysis surfaces documentation holes so your team can fix them before users get frustrated.

Always in sync

Incremental sync keeps your KB current. Checksum-based delta detection means only changed content gets re-embedded — fast, cost-efficient, always fresh.

MCP server built-in

One click to expose your knowledge base as an MCP server. Engineers in Cursor, VS Code, or Claude Code get accurate, cited answers about your product without leaving their editor.

Full observability

Every query traced end-to-end. Token counts, retrieval scores, tool calls, and confidence levels — you see exactly what the agent did and why.

Integrations

Connect every source of truth

50+ plug-and-play connectors. Each one optimised for LLM retrieval — not just a raw text dump.

⚙️GitHubCode
🦊GitLabCode
📘ConfluenceDocs
NotionDocs
🎯JiraIssues
💬SlackComms
🎮DiscordComms
🎫ZendeskSupport
📐OpenAPISpecs
🦕DocusaurusDocs site
📚MkDocsDocs site
📖GitBookDocs site
🗣️DiscourseCommunity
📄File uploadFiles
🌐Web crawlWeb
▶️YouTubeVideo
☁️SalesforceCRM
🔌+ 30 more

Missing a connector? Request it or build one with our open connector SDK.

Deploy everywhere

Meet your team where they already work

Deploy in minutes. Switch channels anytime. One knowledge base, every surface.

🌐

Website Widget

Most popular

Embed a chat widget on your docs site with a single script tag. Fully customisable — your brand colours, your logo, your copy.

<script
  src="https://widget.dysha.ai/embed.js"
  data-project-id="YOUR_PROJECT_ID"
  data-color="#4f46e5"
  data-name="Ask AI"
  async
/>
💬

Slack Bot

Deploy dysha.ai as a Slack bot. Engineers ask questions in any channel or DM — answers arrive in seconds with citations. Reduce the noise in #ask-engineering.

🔌

MCP Server

New

One click to expose your KB as an MCP endpoint. Cursor, Claude Code, VS Code, and ChatGPT can query your docs without leaving the editor.

// .cursor/mcp.json
{
  "mcpServers": {
    "dysha": {
      "url": "https://mcp.dysha.ai/YOUR_PROJECT_ID",
      "headers": { "Authorization": "Bearer YOUR_KEY" }
    }
  }
}
🔧

REST API

Full programmatic access. Integrate dysha.ai into your own products, agentic workflows, CI pipelines, or support tooling.

curl -X POST https://api.dysha.ai/v1/query \
  -H "Authorization: Bearer YOUR_KEY" \
  -d '{ "question": "How does auth work?",
        "project_id": "YOUR_PROJECT_ID" }'

RAG quality

The most reliable retrieval pipeline for technical content

Most RAG systems embed, retrieve, and hope for the best. dysha.ai runs a multi-stage pipeline that verifies evidence before generating a single word — so your users never get a confidently wrong answer.

1

Hybrid retrieval

Dense vector + BM25 keyword search over your indexed chunks, fused with Reciprocal Rank Fusion.

2

Cross-encoder reranking

A cross-encoder jointly scores query + chunk pairs for fine-grained relevance — 20–30% precision improvement over bi-encoder alone.

3

Evidence validation

Retrieved chunks are evaluated for confidence: high / medium / low / insufficient. Low confidence triggers a retry with a refined query.

4

KB-only grounding

If the knowledge base has no relevant content, dysha.ai says so. It never fills gaps with training-data guesses.

Query pipeline trace

queryHow does token refresh work?
searchsearch_kb → 12 candidates found
rerankreranker → top 5 selected
validateconfidence: high · recommendation: answer
answerGenerating cited answer…

Answer: The middleware uses a sliding window strategy — tokens within 5 min of expiry are refreshed automatically. [auth/middleware.ts:142]

Get started with
dysha.ai pricing

dysha.ai pricing is tailored to your specific team's needs.

  • AI platform fee based on your needs incl. optional add-ons
  • Flexible scaled pricing based on answers per month
  • Support and integration with your tools included with every plan

FAQ

Common questions