Currence (formerly Sightline Climate) runs on an AI data engine that ingests thousands of sources into one live view of the companies, projects, deals, and costs that move energy markets. It is what lets a strategy or innovation team see in one place who is building what, where, and at what capacity, and how that picture is changing from one week to the next.
The data we pull in is messy, though, which is why the first thing we have to do is enrich it. Ingestion can bring in a news article or a developer spec sheet, but on its own that is just a document on a page. The enrichment agent reads it and turns it into structured, trustworthy data about a real project, covering the site, the status, the capacity, the power and supply type, and all the other fields you need to make a decision on.
Energy data is everywhere, and none of it agrees
Energy data lives all over the cloud, showing up in news publications, regulatory filings, developer spec sheets, interconnection queues, and a hundred other places, almost always in a different format each time. It also varies a lot by sector. A data center project needs details like its powering model and redundancy tier, while a clean fuels project needs feedstocks and offtake.
We looked hard at whether off-the-shelf tools could solve this, and we found that they couldn't. The deep research agents that are readily available don't understand the taxonomies hand-built by our analysts, and building a custom enrichment pipeline for every sector one at a time would have been slow, expensive, and impossible to scale as we add new data types over time.
What we needed instead was a single flexible architecture that could work across any sector and any source.
How the enrichment agent works
Most of the successful agents out there, including Claude Code, share the same straightforward architecture, which is a while loop that calls tools. Each pass through the loop hands an input to a large language model and asks it to produce the answer you are looking for. If the model can answer from what it already has, it simply returns the result, and if it can't, it looks through its bank of tools for something that would help, calls that tool, and feeds the result back in. It keeps looping through the tools until it has the answer you need.
We took that pattern and added the part that makes it ours, which is the set of research processes our team has tried, tested, and come to trust. The agent starts with input data, which might be an RSS feed, an Excel file, or a spec sheet, and tags it to a sector profile. An article from Data Center Dynamics or a data center developer spec sheet gets tagged to the data center profile, while an interconnection queue spreadsheet gets tagged as a power project. Each profile defines what we care about for that kind of project, including which fields to fill, the extra context the model needs for each one, what to query, and which tools to load into the bank for that run.
Once the agent has the input and the profile in hand, the loop begins. It first tries to find every value the profile asks for, and then it checks whether it actually found them all. If some are still missing, it looks for another tool to run, such as geocoding an address into coordinates, runs that tool, feeds the output back to the model, and keeps going until it has filled in as much of the profile as it can.
Unlocking new sources for energy data
We added one more piece that we are especially proud of, which is the ability to discover new places to look as the agent works. On each pass through the loop, it asks whether the data it just found unlocks new sources worth checking.
A clear example is what happens once we learn that a data center sits in a particular state, because at that point its permitting and regulatory filings suddenly become valuable. We built a tool so that when a project lands in Texas, the agent goes to the TCEQ, which tracks Texas air permits, and searches by project name, developer, and location for anything tied to that data center. Whatever it finds gets ingested, and that can surface still more detail, such as the equipment on site or additional contracts.
When the agent has collected every value and every source it can, it moves to the final step of synthesizing the calculated fields. That includes things like a clean project description, so you can grasp what matters about a project at a glance before digging into the details. What comes out the other end is a project page on the Currence platform with the site, status, capacity, power and supply type, and everything else the agent discovered, all in one place.
Why this matters
Energy market intelligence is only as good as the data behind it. What makes our approach different is that we are feeding the research you already know and trust directly into Currence's agents. The same intelligence behind the analysts you have seen speak at conferences and in webinars now runs inside an automated system, and that system keeps learning from the work they continue to uncover, so the data that comes out is comprehensive, consistent, and traceable.
Our market intelligence maps to how energy teams choose partners, forecast demand and price, and decide where to build, buy, or invest next. More than 90 teams already run on the Currence engine, including Microsoft, bp, Baker Hughes, Southern Company, HSBC, BBVA, Siemens Energy, Shell, BHP, B Capital, Galvanize, and Mitsui.
If you’re interested in learning more about our coverage, request a 20-minute call with our team.
.png)
