In Review

Planned

In Progress

Completed

Main Roadmap

Beehive Climate’s context data are unstructured PDFs. Some of these docs have embedded tables. The content in these tables is not getting vectorized reliably which creates a problem for their application. Aman described a hypothetical node that would handle this use case.

Create node to vectorize structured data within unstructured docs

cwhiteman

It would be helpful when testing a Flow to only see test cases for the selected Flow. One way to do this would be to use the schema to filter out test cases that don’t match the Flow.

Associate Test Cases with Flows

If you’re looking at a list of multiple jobs, it would be helpful to see which vector db each updates so you can toggle on/off the right one without having to go look at each flow: <img src="https://fb-usercontent.fra1.cdn.digitaloceanspaces.com/01942cf0-7da0-799e-805c-797b9796ac72.png" data-width="100%" data-align="left">

Enhance Jobs list by showing the VDB it hits

Ability to add sticky notes inside flow

aruna

do native integration with supabas<ul><li>CRUD database</li><li>CRUD file storage</li><li>Crud users and other supabase API</li></ul>

Supabase integration

Aman Sharma

Dev design a form in sync and asynchronous mode User trigger flow on form submission which can respond

Customisable Form widget

Allow record insertion and fetching from postgres

Insert and fetch postgres

Integration with S3

Display logs related to the deployment in case of deployment failure.<img src="https://fb-usercontent.fra1.cdn.digitaloceanspaces.com/ss-0193ab64-a5d4-72af-ab35-ee5ff7f3d6fd.png" data-width="100%" data-align="left">

Show Deployment Failure Logs

User

Add support for deployment version history, allowing users to view a list of deployed versions and roll back to a previous version if needed.

Deployment Version History and Rollback Support

When creating Test Cases, capture expected results (in addition to capturing the name and test conditions as Lamatic does currently). Consider also,<ul><li>Using code checks and AI to find deviation and grade result quality.</li><li>Use this quality score as an input for deciding whether to deploy an update.</li><li>Assist in recommending the best option when A|B testing prompts, agents, workflows, models and other settings such as temperature.</li><li>Using these results to create benchmarks.</li><li>Measuring and alerting about “model drift” over time.</li><li>Integrating “human-in-the-loop” strategies to expand Test Cases and quality checking.</li></ul>Added Later (based on feedback from another user):<ul><li>Make it easier to see the history of Test Case executions (without leaving Experiments). Marc Greenberg didn’t realize he could find the results in Logs, but also didn’t like having to dig through logs to find the execution details for a given test.</li></ul>

Test Case Expected Results

When chunking, vectorizing and indexing an unstructured document, make it possible to correctly index any structured content embedded in that document.This might leverage logic from the Multi-modal Node. This is needed by <a target="_blank" rel="noopener noreferrer nofollow" class="link" href="andrew@beehiveclimate.com">Andrew</a> at Beehive.

Parse docs that contain structured data (i.e., tables)

Primary Key not understandble - make this selectable from metadata fields instead

High Priority

Primary Key Selection in Indexing node

A dedicated UI to choose model preferences on project level. <ol><li>There will be different categories of models ( Tabs ) for Text | Embedding| Chat | Image | Audio | Transcript </li></ol><ol start="2"><li>For Each category customer can create 5 Settings and reorder them. </li><li>In each settings can choose 4 tabs ( Basic | Fallback | Best | A/B Test ) </li><li>When using model selection in a workflow node, choose a preferred setting options which will populate the UI with credentials and model selection. ( By Default the first preference will be chosen or create a new preferences </li></ol>

Model Preference & Fallback

Airtable Integration

I see users copy/paste content into a test payload and then struggle to remove all the /n line feeds, etc. It would be delightful to automatically convert such content to valid JSON. Andrew at Beehive specifically mentioned this.

Auto-format test data to be valid JSON

Right now we show that we tested the integration but we actually didn’t start doing this for all the integrations

Engineering

Test Credentials after saving

Have a search + chat like ask widget that users can embed on there website example - mintlify docs <img src="https://fb-usercontent.fra1.cdn.digitaloceanspaces.com/01936eb1-9d62-71c2-b59e-62607c2e3edd.png" data-width="100%" data-align="left">

UI/UX

Embeddable Ask widget

Currently we offer limited values for existing search. We should support more values based on these docs - <a target="_blank" rel="noopener noreferrer nofollow" class="link" href="https://weaviate.io/developers/weaviate/search/bm25">https://weaviate.io/developers/weaviate/search/bm25</a> <a target="_blank" rel="noopener noreferrer nofollow" class="link" href="https://weaviate.io/developers/weaviate/search/hybrid">https://weaviate.io/developers/weaviate/search/hybrid</a> params -<ol class="tight" data-tight="true"><li>Show bm25 score instead of cosine similarity</li><li>Search on selected properties only</li><li>Use weights to boost properties</li><li>limit results to groups of similar distances to the query, use the <code>autocut</code></li><li>Filter results</li><li>Explain the search results for hybrid search</li><li>Change the fusion method for hybrid search</li></ol>

Support for additional properties for search.

Cancel Failed Batch

Pass history to AI Json generate as well

atech

Signin with google

Documentation shows differently than how the feature is supposed to perform. Mismatches - <ol><li>Host URL</li><li>Scrap vs Crawl Difference</li><li>HTML Tags</li></ol>

Improve Documentation for Firecrawl

Retry failed batch

Change the options to schedule for Batch triggers to the following - <ul><li>Every 24 hours (most common)</li><li>Every 12 hours</li><li>Every 8 hours</li><li>Every 6 hours</li><li>Every 3 hours</li><li>Every 2 hours</li><li>Every 1 hour</li></ul>

Simplify the Scheduled Options in batch Triggers

When adding Google Drive as a data source, Jason Reichle <a target="_blank" rel="noopener noreferrer nofollow" class="link" href="https://sb.traq365.com/share/1806/a3NHj86CqQy97Dsb5JYt24XeAi85Scw3EPg27TmGz">at 15:28</a> encounters a warning “Google hasn’t verified this app”, but proceeds only with my encouragement. He then explains how he would like to limit granted access to a specific folder (Al Serize on 9/4 had same request).Recommended action Items to increase trust &amp; reduce abandons:<ol><li>Set the user’s expectations before they get the warning.</li><li>Update the email address referenced in the warning from <a target="_blank" rel="noopener noreferrer nofollow" class="link" href="mailto:vansha@lamatic.ai">vansha@lamatic.ai</a> to <a target="_blank" rel="noopener noreferrer nofollow" class="link" href="mailto:admin@lamatic.ai">admin@lamatic.ai</a> (more trusted).</li><li>Get Lamatic verified by Google.</li><li>Limit permissions to something less than the entire Google Drive.</li></ol>#1 - Set Expectations:A simple way to boost trust is to explain the permissions they’ll be asked to grant in advance of the Google prompt (see highlighted area in the screenshot below).<img src="https://fb-usercontent.fra1.cdn.digitaloceanspaces.com/0191b872-174a-72a3-90ed-af594a79c862.png" data-width="100%" data-align="left">Suggested message text (for the highlighted area in the screenshot above):<h2>Add Credentials</h2>Add a unique name to reference this Credential within the workflow builder.Note: Google will ask you to grant <a target="_blank" rel="noopener noreferrer nofollow" class="link" href="http://lamatic.ai">lamatic.ai</a> access to your Google Drive. This access is used strictly by the Lamatic application to allow you to create workflows using the information found on the authorized Google Drive.#2 - Update Lamatic app authorReference a more trusted email address than <a target="_blank" rel="noopener noreferrer nofollow" class="link" href="mailto:vansha@lamatic.ai">vansha@lamatic.ai</a> on the warning below (e.g., <a target="_blank" rel="noopener noreferrer nofollow" class="link" href="mailto:admin@lamatic.ai">admin@lamatic.ai</a>).<img src="https://fb-usercontent.fra1.cdn.digitaloceanspaces.com/0191b86a-4ba0-7543-ace1-27c4e7c4e5b6.png" data-width="100%" data-align="left">#3 - Get Lamatic verified by Google:Not sure what is required to do this, but getting verified would increase trust a lot.#4 - Narrow the permissions:Not sure if this is even possible, but the screen below is scary. Multiple users have requested that permissions be limited here to a specific folder (they have sensitive stuff in other folders and don’t want to grant access to this data).<img src="https://fb-usercontent.fra1.cdn.digitaloceanspaces.com/0191b86b-06a3-735c-afe4-51a9b25bd8a0.png" data-width="100%" data-align="left">

Google Drive: set expectations, verify app, and narrow permissions

Show Variable Selector in the API response nodes variables

On failure of any node return the exact point of failure

Create a Web Crawl Node to allow Lamatic to ingest and store data from a website to be accessed downstream. Specific requirements: 1) Accept a URL, multiple URLs, a sitemap.xml, or a URL pattern regex as a parameter to direct the crawl.<ul><li>Consider what would be needed to accept login credentials as a param to enable ingesting content behind a login (future).</li></ul>2) Support the ability to store important metadata along with the visible content (for instance, the hreflang tag that will indicate the language &amp; geography of the target audience).<ul><li>Consider a default metadata capture config that can be optionally reduced or expanded (future).</li></ul>3) Generate status and error messaging:<ul><li>Accept parameters that configure status notifications (what notifications to send &amp; to whom?).</li></ul>Thoughts/Questions: <ul><li>How to tell the Node where to store the crawled content?</li><li>How should the visibility of this store be determined (in other words, is this store available only to this workflow, others within the Project, or others within the Organization)?</li><li>Should the Node support an option to simultaneously vectorize the content as part of the crawl (to reduce storage needs)?</li><li>Can/should we use the &lt;lastmod&gt; attribute on the Sitemap.xml to trigger this Node?</li><li>Can/should the Node capture Image, Audio, and/or Video content?</li><li>How should notifications &amp; error handling work (regarding the status of the crawl)?</li></ul>

Web Crawl Node

View different templates in one Place

Template Library

Beehive Climate is a startup &amp; customer. They’ve created an MVP indexing docs from Google Drive, but their goal is to put this content in S3. They would like to migrate this data to S3 during January.

Add S3 data integration node

Help Center

Improved

Fixed

Integrations

lamatic

Rejected

Low Priority

Tell us how we could make the product more useful to you. For Support reach out at hello@lamatic.ai.

Lamatic.ai

Main Roadmap