Product · MnemoRAG
MnemoRAG, the Mnemo RAG Tool, gives your app, bot or workflow answers grounded in your own data through a single call. There is no search system for you to build, host or tune.
What it is
You define an endpoint over exactly the subjects you choose, including subjects other tenants have shared with you. Your application sends a question and gets back an answer grounded in only that data, with the source, confidence and subjects used. There is no chat session to track.
Why not build your own
Building your own means cutting your documents into pieces, turning each piece into numbers, storing them in a special database, and then searching for the pieces that look most similar to each question. You also have to keep all of it up to date as your data changes. It works for loose text, but similar is not the same as right. The closest match can be the wrong one, a table can be split in half, and the AI is left to add up the figures itself.
1 · Filed, not scattered
M.N.E.M.O. organises everything you store into a structure called an Omni Axis Tree, or OAT. Think of a well-labelled filing system where every fact has a known place and can be reached by several routes, such as by product, by region or by quarter.
2 · Find the drawer, then open it
The AI reads a short table of contents to decide where the answer should be. M.N.E.M.O. then opens that exact place and returns what is stored there, whole, not a similar-looking piece. The answer is written from what was actually retrieved.
3 · Calculated, not guessed
Totals, averages, counts, highs and lows are calculated by M.N.E.M.O. from the stored records. The AI never adds figures up in its head, so a number in an answer is a number you stored or one computed from them.
| Build your own | MnemoRAG | |
|---|---|---|
| What you set up | A way to cut documents into pieces, a model to convert them, a database to hold them, and the code that ties it together | Choose your subjects and get an endpoint to call |
| Finding the answer | Looks for the pieces that seem most similar to the question, so the nearest match can be partial or wrong | Goes to the place the information is filed and returns what is stored there |
| Tables and numbers | A table can be split across pieces, and the AI adds up the figures itself | Rows stay together, and totals are calculated by M.N.E.M.O. |
| When your data changes | Reprocess and retune | New information is filed as it arrives, so there is nothing to rebuild |
| Who can see what | Rules you write and maintain yourself | An endpoint can only see the subjects you list |
| Knowing why | A similarity score | Each answer says where it came from, how confident it is, and which subjects it used |
To be clear about the AI's role: it still decides where to look, using the table of contents. What it does not do is guess at the stored data or do the arithmetic. Those come from the memory.
How it works
What a call looks like
POST /mnemo/rag/<your-endpoint-id>
X-Mnemo-Api-Token: <your API token>
{ "query": "What was total revenue for Asia Pacific in Q1?" }
→ {
"answer": "Asia Pacific revenue in Q1 was 1,042,318.36.",
"found": true,
"source": "Sales 2026",
"confidence": "high",
"subjectsUsed": ["Sales 2026"]
}The endpoint URL and token come from Dorothy Settings. Available on Pro and Enterprise.
In real life
Illustrative scenarios, not customer stories.
A helpdesk calls one endpoint scoped to the product documentation and the resolved-ticket history. A customer asks why an export failed, and the widget answers from the real fix that worked last time, citing the ticket.
A bot calls an endpoint over the sales subject. "What was Q3 revenue for EMEA?" comes back as the computed total, not a language-model estimate, so a number pasted into a board deck is the stored number.
An HR assistant gets an endpoint over the policy subject alone. Payroll lives in another subject that the endpoint is never given, so the assistant cannot answer from it, whatever anyone asks.
A retailer subscribes to a supplier's product-specification subject and exposes an endpoint over it in their own ordering tool. When the supplier updates a spec, the next call reflects it, and if the supplier revokes access, the next call stops seeing it.
For developers
MnemoRAG sits alongside the REST API, the MCP server and the Claude Code Skill.Same memory, other doors
Whatever this feature can see, the other two can see too, so you can add them later without re-uploading anything.
Ask your own knowledge in plain language.
Learn more →Publish a subject; others subscribe and ask.
Learn more →Next step
Choose a plan, or talk with Panaxon AI about your use case.