Tasq v6 Beta Release
We have been quietly working on something big. On December 7th, we will release Tasq v6 (beta) and it redefines what agentic truly means for time series based operations.
The reason we went down agentic:
- Integration is easier for clients
- Accuracy of prompt results improves
- Tasq's "Build your own model" as an agent is the path we want to dive further into and what has taken off as of late
The basis of this release is improving 3 core areas of the business. Current client satisfaction, enabling company growth and building more around our core differentiator.
1. Integration
Agentic means easier and quicker "no hands" integration. Integrated right into Teams, enables low barrier use and reduces onboarding friction. This means we are simplifying user workflows vs the current applications in use. We have made it so easy that users do not need any touchpoints before onboarding, everything is automated to give you insights on the first minute of onboarding.
Connect to db > Tasq Agent Spun up > @Tasq in Teams to give you AI instantly at your fingertips.
This is the future of trials, and will enable quicker iteration for the client. Technology groups are now enabled. Before, technology groups within Oil and Gas were a weight on the business unit, offering a roundtable of new technology that might not solve the end users current painpoints. Now, trials can be 30 minutes with technology groups offering up the "finalized technology" to the end users, making that group more useful, and enabling higher iteration. 3 to 6 month trials are a thing of the past. Integration is no longer a barrier for iterating with new technology.
2. Accuracy of Prompt Results Improves
For the past year our Tasq v5 AI Data Analyst has been a key feature that gets a lot of use. Type anything into it, and Tasq automatically generates code to retrieve the necessary information. We have noticed a lot of use cases from it, but we have also noticed a higher than desired false positive rate from it.
One shot AI was tough to nail down. Time series data is not static text, it is live, contextual, and interconnected. Most AI is missing the mark with time series accuracy, mainly due to context, systems just cannot reason across it.
The core issues we identified with the legacy approach:
- Latency limits and avoiding timeouts due to poorly optimized queries
- Db specific syntax leading to erroneous code
- Added context bloat to the training data
- Post code/results return, there was still follow up from the user to tweak the results
- Reliability in the same prompt varied even with proper training data
Over the past six months, we have been testing Tasq's fully agentic system side by side with legacy code gen SQL prompt versions. The results are a massive improvement. This coordination with Tasq's "Signal Search" is the breakthrough. It is why our accuracy jumped from 50% to 80% with an 8x wider case coverage. The chain of thought, context engineering, memory and agents all thinking together before answering. They validate, test, and refine together. You are not getting a hasty SQL hallucination, you are getting transformation without having to do anything more.
3. Build Your Own Model as an Agent
Agentic means: Building the best in class capabilities for searching time series data. We have also expanded the capabilities of how this works with other Agents.
- Multiple agents that work together to bring the right results embedding the right context in each prompt
- Agents that invoke models, graphs, maps, and recommendations, not just code
- Agents trained specifically on oil and gas models, physics based models, statistical inference, signal search, pattern matching all working together
Searching signals and patterns is transformational for users to build their own models. An agentic system built around this is very powerful. It enables better search, better workflows, better recommendations and everything to be more accurate.
Summary
There is so much more to time series search than SQL. Time series search has not been solved at scale and is the unlock to most workflows in operations. Platforms like Pi, Snowflake and such do not have the capabilities, context, and ability to create models like Tasq does. On the other hand, platforms like Palantir are great data connectors and workflows, but lack the industry intelligence that Tasq offers.
For time series data whether it is to build a model, analyze events, pattern match, summarize performance, assign tasks, or automate workflows, Tasq v6 does it faster and with more intelligence than anyone else.