In May, we released a feature every day. We did the same in January and it led to loads of product improvements. Thank you to all of our users for the feedback over the past month. We have a long list of things we want to build, and your feedback has been invaluable in helping us prioritise the right things.

A big one was launching our new Free plan with API access and database integrations supported. Sign up for free today. We also launched an 8-week program to help you pilot Defog in your app and evaluate its impact.

Both these plans make it easier for users to experiment with Defog and understand its capabilities - without having to invest in a full implementation.

Read on for a summary of what we improved in May.

Upgraded Free Plan

We launched a new free plan with access to our API and database integrations. We are making it as easy as possible for you to explore Defog, build an MVP and demo it to your team. You get 100 API calls per month, on a schema with at most 4 tables and 25 total columns.

This does still run on a less capable model than the ones used by our Pro and Enterprise customers, but is good enough to test out simple queries.

Get your API key for free.

Integrate in <5 minutes

You can now go from pip install to your first query on Defog in less than 5 minutes. Some of the improvements we made to enable this:

A CLI wizard to walk you through the entire process

Automated recognition of indexes and foreign keys

Check if your database is suitable for Defog

Docs

Richer chat UI with built-in visualisations

Defog’s front-end components now have much richer features. Improvements include:

A new chat interface that retains context from previous questions. It is also installable with npm i defog-components

Automatically convert questions and tables into charts with a couple of clicks

A summary of the main insights generated from the resulting query

Websockets to show status updates to users while a query is running

Option to enable dark mode

Docs

Give feedback with Guided Teaching mode

If a query isn’t producing optimal results, you now have the ability to give the model feedback. This will lead to much quicker improvements in query accuracy when you deploy Defog.

Debug mode for your internal beta testers to give a 👎 or 👍 to results

Guided teaching to give the model detailed feedback on what isn’t working

Add definitions to your glossary of terms at the user-level to teach Defog about very specific terms and what they mean in your business context

Deploy Defog on-prem (beta)

You can now deploy Defog on-prem, using models fine-tuned on your schema. This is currently available to enterprise customers

Multilingual & Personality

Defog can now understand questions, ingest schema and provide answers in multiple languages. We added support for: Arabic, Mandarin, Spanish, Hindi, Portuguese, Korean, Japanese and Bahasa Indonesia. You can now also specify a personality for Defog in your product.

Statistical modelling (beta)

Until recently, Defog could only generate SQL queries to answer data questions. Now, it can also create Python code and run regression models. Here’s a <2 min demo.

Reporting agent (beta)

Our new reporting agent can handle broad questions like “how can I increase revenue for my business?”. It analyses multiple scenarios and hypothesis and provides a set of recommendations based on the data.

MedSQL model

We launched a dedicated model for healthcare customers. The MedSQL model lets you query US healthcare billing data in the ICD-10, CPT and HCPCS codes in natural language. Like the rest of Defog, this works in a privacy-friendly way and ensures that your data never leaves your servers.

Question: how many patients did a screening for breast cancer?

import os
import requests
r = requests.post("https://api.defog.ai/generate_query_chat",
    json= {
        "api_key": os. environ['MED_API_KEY'],
        "question": "how many patients did a screening for breast cancer?",
        "db_type": "postgres"
    }
)

Result:

SELECT COUNT(DISTINCT beneficiaries.id) AS num_patients
FROM claim_items
JOIN claims ON claim_items.claim_id = claims.id
JOIN beneficiaries ON claims.beneficiary_id = beneficiaries.id
WHERE (
    claim_items.cpt_code = '77063'
    OR claim_items.cpt_code = '77067'
    OR claim_items.hcpcs_code = 'G0202'
)

Pilot Program

We launched an 8-week program to help you pilot an AI assistant in your app, gather feedback from users, and evaluate its impact on your business.

More details on our Pilot Program.

That’s all the updates from May 2023. We’ll be back with our next update here in a month. We release features and updates every week - follow us on Twitter or LinkedIn for weekly updates.

← More blogs

May 2023 Product Update