Skip to content
Stedle AI

Own product · open-source platform

Knowledge Agent Platform

A multi-tenant platform that turns a company’s documents into a cited chat and voice assistant, embeddable on any website with one script tag.

AI Agents & AssistantsAI Integration & RAG EngineeringAI Product DevelopmentKnowledge-heavy teams
Redrawn product screen. Names, businesses, people and prices are fictional sample data.

Verified in the code

automated backend tests
978automated backend tests
end-to-end browser tests
21end-to-end browser tests
API operations
93API operations
embeddable widget
~10 KBembeddable widget

The problem

Companies want an assistant that answers from their own policies, catalogues and manuals, but generic chatbots guess, mix data between customers and can’t show where an answer came from.

What we built

We built a full platform: upload documents, and the system extracts, chunks and indexes them, then serves a grounded assistant through chat and push-to-talk voice. Every answer cites its source, and each company’s data is isolated at the database level.

Capabilities

What it does

  • Answers only from uploaded documents (PDF, Word, Excel, PowerPoint, web pages) and cites its sources
  • Chat assistant plus push-to-talk voice: speech-to-text, streamed spoken replies, interruptible mid-answer
  • Embeds on any website with one script tag via a ~10 KB widget, themeable to the host brand
  • Hybrid search: vector similarity plus full-text, merged with Reciprocal Rank Fusion
  • A chunking agent picks one of five strategies per document and explains its choice
  • Tenant isolation enforced by Postgres row-level security, with dedicated isolation tests
  • Uploads are virus-scanned, extracted, de-duplicated, quality-checked and indexed in background workers
  • Admin dashboard: knowledge bases, document pipeline status, assistant testing, embed code, members, API keys, analytics and audit log
  • MFA, single sign-on, API keys, ten built-in roles and prompt-injection delimiting

Under the hood

How a question is answered

  1. 1Question
  2. 2Intent analysis
  3. 3Plan
  4. 4Hybrid retrieval
  5. 5Grounded answer
  6. 6Citations
  7. 7Chat or voice reply

Stack

PythonFastAPIPostgreSQL + pgvectorRedisCeleryLangGraphOpenAIAnthropicWhisperNext.jsTypeScriptDockerPlaywright
Next projectAI Front Desk for Dental ClinicsView case study

Want something like this for your business?

Book a free call. We’ll look at how your business runs and tell you honestly what’s worth automating, and what isn’t.

Book a callSend a message