Overview
Exomind (ExoInsights) = RAG service for indexing/training documents & websites and returning chunks to the LLM. It does not replace intent/NLP; it supplements when intents fail.
Bot Mandate = backend (Django) toggle that routes requests to the mandate/AI agent instead of the default NLP. Must be enabled from DB/Django.
Gen-AI Orchestrator = enables combined flow: NLP for intent-driven flows, Bot Mandate + Exomind for FAQ/data retrieval. Requires both orchestrator flag and bot mandate enabled.
How to Enable Exomind
A - Quick UI steps
For (v8 UI)
Open the Bot → click Edit (bot settings).
In the left navigation, find Solution Settings.
Locate Enable PDF Searcher → turn it ON.
A dropdown appears below — select Exomind (beta).
Optional: enable Show source link — when enabled, bot answers will include links to source documents/websites.
Optional: Enable Exomind custom prompt — add your system instruction here (this will be sent as a system prompt when Exomind content is used).
For (v7 / old UI)
Open Bot → Edit → locate Enable PDF Searcher (no tabs).
Toggle it ON and choose Exomind (beta) if a dropdown exists.
Same optional settings (source link, custom prompt) apply.
B - Train documents/websites in Exomind (v8 workflow)
In v8 side nav, go to Workflow.
On the right side, click the Exomind (beta) tab.
Add documents or websites - upload PDFs, point to URLs, or paste content.
After adding, Start Training (click the checkbox/train button).
Wait for the training result - status will show Completed or Failed.
If completed, the trained chunks are available to be fetched by the bot.
If failed, check logs / re-upload and retry.
C - How Exomind is used at runtime (behavior)
User query → NLP / intent matching is attempted first.
If intent matched → NLP reply (no Exomind).
If no match → request falls back to Exomind (Gen-AI/RAG).
Exomind returns matched chunks from trained sources. Those chunks + (custom or default) prompt → sent to LLM to generate an answer.
If LLM returns an answer, it is shown (optionally with source links). If not, a failure response is returned.
D - Verify Exomind is working
Train a small doc with a unique sentence: e.g - “EXOMIND_TEST_XXXXX” and mark it in the document.
Ask the bot a question that will not match any intent but is answered by that sentence.
If Exomind is enabled & training completed, you should get the answer (and a source link if enabled).
E - Troubleshooting tips
Training stuck/failed → recheck document format (PDF text vs scanned image), re-upload as text or OCR first.
No chunks returned → confirm project+bot id mapping in Exomind Bot API details (account_id & project_id).
Answers look irrelevant → consider editing or enabling a custom prompt to shape how chunks are used.
How to Enable GenAI
Important: GenAI (Bot Mandate) can only be enabled from backend by the admin.
This is a safety/restriction: Once enabled, the mandate configuration will control routing.
To enable this feature, please reach out to your respective Account Managers or raise an email to help@exotel.com.
Please provide the required details in the email to avoid to & fro. Subject as "Enable Gen AI for bot_name".
A - How will Bot work after enabling GenAI
The bot will route requests directly to the Bot mandate (AI agent) - NLP will not run for that bot.
Configure the mandate from the AI Agent UI (side navbar) - set flows, add Exomind config if needed.
Ensure the mandate config maps to the Exomind project (API call) if you want it to use Exomind chunks.
C - Create Gen AI Bot Steps
Please go through this detailed document to understand and create a Gen AI Bot - Create a Gen AI Bot
D - Notes & Cautions
Only the config defined in the mandate will be used; if the mandate config is incomplete, the bot might return unexpected responses or failures.
Test after enabling with a known intent and a fallback query:
Known intent should be handled by the mandate config if it has the flow (otherwise, fallback behaviors depend on how the mandate is implemented).
To Disable the mandate: Raise it to your respective Account Managers
NLP to Gen-AI Switch (Beta)
A - Purpose
Goal: Combine NLP intent flows and Gen-AI so that:
Intent-driven flows (e.g., transactions, guided steps) are handled by NLP.
FAQs/knowledge lookup from trained docs/websites are handled by Bot Mandate + Exomind.
Example:
User asks: “What are the account savings options?” → GenAI fetches docs and answers.
User asks: “I want to apply for a loan.” → NLP recognizes intent and triggers the loan application flow.
B - Enable steps (Django + Agent UI)
This needs to be raised to your respective Account Managers.
C - How runtime works (user journey)
User sends a message.
System checks: Is NLP-Gen-AI Switch enabled?
If yes: The switch manages the logic of switching between NLP and GenAI as per configurations.
Typical logic:
Try NLP first for intent match.
If NLP success → NLP response or triggers a flow.
If NLP fails → fall back to Bot Mandate.
Mandate calls Exomind to fetch matching chunks (POST to Exomind matchedchunks endpoint).
The LLM forms the answer using chunks and the configured prompt.
If the user later triggers NLP (e.g., clicks a suggestion or triggers a flow), the orchestrator will route to NLP for subsequent queries.
D - Limitations of this Switch:
Once exited, flows cannot be resumed.
During NLP flows, FAQs and document-based queries are blocked.
If both NLP and LLM fail, the answer depends entirely on the Content trained in Exomind.
GenAI answers are based on available trained content (PDFs + website data) and Gen-AI free flows. Untrained topics will fall back to a failure message.
Flow switching allowed only between predefined intents.
No mid-flow GenAI handover is permitted.
Quick FAQ
General FAQs
Q: What is Exomind?
A: Exomind is a Retrieval-Augmented Generation (RAG) service that lets you train your bot on documents or websites and use them to answer user queries intelligently.
Q: Can I train both documents and websites in Exomind?
A: Yes, Exomind supports training from both PDFs and web URLs. You can upload multiple sources under one bot’s project.
Q: How long does Exomind training take?
A: It depends on the file size and number of pages. Small documents (1–5 MB) usually complete training in a few minutes.
Q: Can I retrain or update existing documents?
A: Yes. You can re-upload updated documents or delete the old ones and retrain.
Q: What happens if two documents have overlapping information?
A: The LLM may pick the most relevant chunk based on the query context. You can refine it using a custom prompt to prioritize one document over another.
Bot Mandate FAQs
Q: Why is Bot Mandate disabled by default?
A: Because mandate routing changes how queries are processed. It bypasses NLP and directly uses the AI agent, so it’s only enabled when the configuration is ready. It is better to discuss the complete requirement with you Account Managers and plan the next steps.
Q: Do I need Exomind for the GenAI to work?
A: Not necessarily. GenAI can work with its own configuration, but Exomind adds document-based answering capability. Whereas, without Document/website training, the Bot will respond using the details available on the Internet and hallucinate.
Q: What happens if the config is missing?
A: The bot will not respond properly because the routing will not find any mandate instructions to process the user query. - Check with your Account Managers or raise a ticket to help@exotel.com
Gen-AI Orchestrator FAQs
Q: Is Gen-AI Orchestrator required for Exomind?
A: No. Exomind can work independently as a fallback. Orchestrator is only needed when you want both NLP and Gen-AI to work together.
Q: What happens if NLP and GenAI both detect intents?
A: NLP is prioritized. If NLP fails, the orchestrator allows GenAI to interpret and answer queries.
Q: Why is Bot Mandate mandatory for Orchestrator?
A: Because Orchestrator relies on Mandate to handle the Gen-AI side (RAG/Exomind flows). Without a Mandate, it cannot route to Exomind or other AI configs.
Technical / Integration FAQs
Q: What should I do if training fails repeatedly?
A: Check document encoding. If it’s a scanned image, use OCR first. Also, ensure your internet connection and upload size are within limits.
Q: Does Exomind store confidential data?
A: Exomind indexes the document for search and retrieval. It’s recommended to upload only data that complies with your organization’s privacy and security policy.
Q: Can I enable multiple Gen-AI orchestrators for the same bot?
A: No. Each bot can have only one orchestrator configuration active at a time.
Q: Does Exomind affect bot speed?
A: Slightly — since it makes an additional call to retrieve chunks and generate responses. However, caching mechanisms reduce some delays.