As of Jan 15th, we've extended our list of supported file types that Agent's can handle (both in the agent's knowledge as well as uploaded directly in the conversation). Agents can now also read any "text" file—meaning files with mime type text/*. This includes but is not limited to: .csv, .htm, .html, and .py files.
View the full list of supported files here: https://docs.ninjacat.io/docs/adding-data
First, some context: Files uploaded to an Agent's Knowledge are only inserted into conversations / the context window when needed. Which is good, so that we avoid burning through the limited context window just from all the uploaded files, but we want to make sure that files are used when they should be. Previously, Agents only had filenames to decide which files to use, which sometimes led to missing or incorrect file selection.
New Feature: File Descriptions are now auto-generated for all newly uploaded files (as of the release a few hours ago). Agents now use both filename and description to determine when to insert files into conversations. This should help ensure the right files are used at the right time.
Note: Files uploaded before this release will have a description field available, but it must be manually added.
More info: https://docs.ninjacat.io/docs/adding-data
For the first time, we're now offering models from Google as model options for Agents: Gemini 3 Pro and Gemini 3 Flash.
Why these models are a big deal: -Massive 1M Context Window: While our current Anthropic models are great for reasoning, Gemini's 1-million-token context window is a game-changer. For those scenarios where you're consistently running into context window max, try switching to a Gemini model. -Low Cost Profile: These models offer a significantly lower price point, especially for high-volume tasks. Gemini 3 Flash is particularly optimized for scale, coming in at just $0.50 per 1M tokens—making it our most cost-effective option for speed and scale.
Our Model Selection documentation has been updated with this information here: https://docs.ninjacat.io/update/docs/model-selection
You can now programmatically monitor dataset ingestion health using the NinjaCat Management API. This feature provides visibility into when dataset ingestions fail, return no data, or are delayed, helping teams identify and respond to issues before they impact reporting or users.
With the new Ingestion History / Batches endpoints, you can:
This capability is designed to support a variety of workflows, including scheduled health checks, custom monitoring scripts, and automated alerts using various tools. Teams can choose how frequently to check ingestion status and which datasets are most important to monitor.
To get started, provide your Management API credentials and call the ingestion history endpoints with supported filters (for example, checking the last 24 hours for errors or no-data responses). When issues are identified, remediation—such as fixing credentials, updating permissions, or running backfills—is handled directly in the NinjaCat UI.
Documentation can be found here: https://documenter.getpostman.com/view/4308783/SVSGMpwr#91ec4a09-c561-4287-bdf2-c14988c9412b
The current Campaign Manager 360 API version has been upgraded to the latest API version, V5 to ensure continued functionality and data availability for our customers.
At the chat-level, you are now able to view how much time was spent, how many tokens used and cost per individual tool call and assistant message. This is just the first step to bringing some usage visibility to the UI so that our users have more intel about how much their conversations with agents are costing them.
This information can be seen in standard conversations with Agents, as well as when working w/ Agent Builder Bob and in the Agent Builder Chat Preview.
We have updated our Agent-Level Permissions to allow more flexibility when granting users chat and edit access to Agents.
The Access settings have been moved to a new "Settings" tab in the Agent Builder where both an Organization Access can be set (Can Edit, Can Chat and Private), as well as the option to give certain users a different level of access. View all the access rules and how to manage permissions here: https://docs.ninjacat.io/docs/managing-access
This now allows users the flexibility to keep their perfected Agents as chat only for everyone in the org, except for those peers that you also want to be able to manage the settings of that Agent (or vice versa).
The current Criteo API version has been upgraded to the latest API version, 2025.10 to ensure continued functionality and data availability for our customers.
We've rebuilt our ad preview system to fix the persistent issues with broken and expired Facebook ad images in your reports and dashboards.
What's New:
We've made several enhancements to our Data Apps feature, improving load times and security:
We have consolidated the way we preview and deploy Data Apps, resulting in faster load times when saving edits to your Data App. In some cases this could take over a minute, and now this takes just a couple of seconds.
Data Apps are now more secure so that you must be a NinjaCat user in order to open up a Data App (that has not been made public). See third point here...
If you are interested in sharing a Data App with a non-NinjaCat user, you still have the ability to do so, but we've just made it a more formal and secure method of sharing. There is now an option in the Configure tab of the Data App Builder to "Enable Public URL." This generates a public URL so that once the Data App is saved, this URL can be shared with others (so anyone with the link is able to access your Data App). You also have the option of later disabling this toggle. In doing so, the Data App results in an 'app not found' for any user that attempts to open the link.
