Products
Prism: making big data understandable
What Prism is
Prism is Blacksky's data platform, and with AgentOS one of its two biggest products. It takes a large, scattered body of information, from official sources, archives, feeds or an organization's own systems, and makes it understandable: collected, structured, enriched, connected and searchable by meaning.
Prism is what Blacksky's agents know. AgentOS agents read from Prism to answer people's questions, and Prism's data can also power websites, dashboards, datasets and APIs directly.
Prism at scale: KnowGov
KnowGov, Blacksky's free, nonpartisan civic platform, runs on Prism. Its Prism holds more than 11.9 million documents: about 131,000 bills with full text and every step of their journey, 3,800 officials from Congress to governors, mayors and state legislators, voting records, elections and candidates, polls, executive orders, court cases and news from across the spectrum. It grows every day through automated intake from official sources. Shirley, a civic agent on AgentOS, answers questions from it.
What Prism does
- Collects. Intake engines pull material from official sources, archives and feeds on a schedule, and every record keeps where it came from.
- Structures. Records become things that can be queried: an official, a bill, a vote, a committee, an issue, a story.
- Enriches and connects. Records are tagged and linked: a bill to its committees and votes, an official to their record, a story to the issue it is about. Language models add summaries and tags where they help.
- Makes it searchable by meaning. Material is indexed so the right passage turns up even when the question uses different words.
- Publishes. A daily publishing step decides what is shown, for example composing balanced news from several outlets rather than one side.
The tiers
Prism is organized in tiers, so an organization pays only for what its work needs:
- Raw records: material as it came in, kept as the record.
- Structured records: the facts pulled out of that material, organized to be queried.
- Search by meaning: material indexed for retrieval, for agents and site search.
- Cold storage: older material kept cheaply for the long term, with snapshots.
A simple phone agent needs little. An agent that remembers returning callers uses more. A platform built on millions of public records uses all of it.
Principles
- Clean provenance. Every record keeps its source. Prism holds only data that can be traced and trusted.
- Separate what should be separate. State records stay separate from federal ones; one client's Prism is separate from every other client's.
- Read without changing. Agents read from Prism; they do not rewrite the record.
- Coverage over perfection. Take what can be had reliably and factually, starting where the most people are, and fill in the rest over time.
One Prism, many uses
A single Prism can serve several agents, a website, a dashboard and an API at once, each with access to what it needs. An organization can connect agents to an existing Prism, or have a new one set up as part of its stack on Black Cloud.
Data and ownership
A client's Prism is part of their dedicated stack, separate from every other client. The data belongs to the client and is exportable. Prism tiers appear as their own lines on the Black Cloud bill.
Common questions
Can Prism hold our existing data? Yes. Bringing an organization's records into Prism, and keeping them up to date, is part of the build.
Do we need all the tiers? No. Most agents start with structured records and search, and add tiers as they grow.
Can we get the data out? Yes, as datasets or through an API, and it is always exportable.