Memory Hub

Manage your knowledge bases in one place.

Get Started

The Memory Hub is a central console to store and organize data for your agents. Embed data from almost any format. Content is chunked and indexed, with optional tags and rankings. Convert structured data into queryable tables. Create multiple knowledge bases. Manage self-generated reinforcement memory.

Inspect every chunk, token and retrieval.

Features

Free Text

Write knowledge in directly, or start from a template: FAQs, objections and rebuttals, and word and phrase rules. Each entry is saved as structured markdown, so the agent retrieves it cleanly.

Add Text Memory with the Objection / Rebuttal template selected from Free Text, FAQ, Objection / Rebuttal and Word & Phrase Rules, a roofing quote objection and how to respond, and the entry saved as markdown

Websites

Paste a URL and Myna maps the site, pre-selecting the homepage and high-value pages while leaving blogs and legal pages out. Choose as many pages as you need and Myna scrapes them and strips out the HTML noise, keeping just the content. Recrawl on a schedule so answers stay current.

Select Pages to Import for a roofing contractor's site with 110 pages found and 15 selected, a scraping progress card with finished pages and their chunks, and a weekly recrawl schedule

Documents

Upload PDFs, Word docs, spreadsheets, CSVs, text files and images. Myna extracts the content, cleans out headers, footers and page numbers, and organizes it into chunked, indexed memories.

Upload Documents with a warranty PDF, an insurance claims Word doc and a zip code spreadsheet being processed, and the PDF shown extracted, cleaned, chunked and indexed into six chunks

Structured Data

When Myna detects structured data, like a price list or a service-area sheet, it's stored as a typed table instead of text chunks. Agents look up the exact row they need, so prices, specs and zip codes come back deterministically and accurately.

A roofing price spreadsheet detected as a table with a typed schema, and an agent answering a metal roof question with an exact table lookup returning $910 per square and a 40-year warranty

Retrieval Analytics

See how often your agents pull from memory over time, which embeddings are retrieved the most, and how much you've stored across your knowledge bases.

Memory Hub dashboard with 142 memories, 486 chunks and 198k tokens, 8.4k retrievals over the last 30 days, the top retrieved breakdown, and General Memory cards for a website import, an objection, an FAQ and a policy

Building a good knowledge base for your agents

Read the guide
  • Answer real customer questions
  • One topic per memory
  • Import pages, not whole sites
  • Put prices in tables
  • Scope knowledge by location
  • Write rebuttals for common objections
  • Tag everything
  • Trim marketing fluff
  • Correct replies in the Playground
  • Recrawl when your site changes
  • Set words to use and avoid
  • Watch what gets retrieved

Questions about Memory Hub

Something else on your mind? The documentation has the details.

Read the documentation
What can I add to Memory Hub?

Free text and templates (FAQs, objections and rebuttals, word and phrase rules), websites, and documents: PDF, DOCX, TXT, CSV, XLS, XLSX and images such as JPG, PNG, GIF, BMP and WEBP, up to 10MB each.

How does website import work?

Paste a URL and Myna maps the site, pre-selecting the homepage and high-value pages. Choose as many pages as you want, and Myna scrapes them, strips out the HTML noise and keeps the content. Pages you've already imported are skipped, and you can recrawl on a schedule.

What happens to large documents?

Content is cleaned, then split into chunks and indexed. You can open any memory to see its chunks and their token counts.

How is structured data handled?

When Myna detects structured data, like a price list or service-area sheet, it's stored as a table. Agents look up the exact row they need instead of searching text, so numbers come back exactly as written.

Can I keep knowledge separate for different agents?

Yes. Group content into knowledge bases, for example one per location, and turn each on only for the agents and workflows that need it. General Memory is shared across your workspace.

How do agents decide what to retrieve?

Agents search memory when the conversation needs it. Results are ranked, and only those above a relevance threshold make it into the answer, so weak matches don't pad the reply.

What is feedback memory?

When you save a correction or reinforcement on a reply in the Playground, it's stored as feedback. Once approved, agents with Feedback Memory enabled retrieve it automatically on later turns.

Can I see which memories my agents use?

Yes. Memory Hub shows retrieval volume over 7, 30 or 90 days, your most retrieved memories, and totals for memories, chunks, tokens and storage.

Start building for free

Launch your first agent in minutes.

Sign up free