Products
AgentOS: the platform Blacksky's agents run on
What AgentOS is
AgentOS is the platform behind every Blacksky agent. An agent on AgentOS is a persona, a set of documents, and a configuration. The platform supplies everything an agent needs to work with real people: conversation, retrieval over documents with sources, memory, a decision tree for common paths, actions that a person approves, and dashboards for the people who run it.
New agents are composed from these parts rather than built from scratch, which is why a new agent can go from idea to working demo quickly.
Agents on AgentOS today
- Maurice, Blacksky's own AI, on the phone and on the web.
- Shirley, a civic agent that answers questions about government from public records.
- A personal agent that runs entirely on an iPhone, with no cloud at all.
Ways to meet an agent
An agent on AgentOS can be present with people in different ways, and each is switched on or off per agent:
- On the phone (Dewey). Fast, spoken, short answers, interruption, keypad menus, handoff to a person.
- In a web chat. Written answers with room for more: the documents behind each answer, cards for related work, suggested next questions, and a recap the visitor can save.
Same agent, same documents, same rules. Only the way it talks changes.
What makes it different
Answers trace back to documents. Agents answer from a curated set of documents and can show which ones they used. When the documents do not cover a question, the agent says so, and the question is logged as a gap to fill.
People approve what agents promise. Agents can commit to things, like a callback or an email, but only the things they are given, and a person confirms each one. An email an agent promises is drafted for a person to read, edit and send.
Agents can reach the operator. When an agent commits to something or runs into trouble, it emails the person who runs it. Nothing goes to anyone else without approval.
Model agnostic. AgentOS works with hosted open models, local models, and models running on the device itself. An organization is not locked to one provider, and the model can change without rebuilding the agent. GoodBrains, Blacksky's model foundation, offers that choice through one API.
Costs are visible. Every model call is recorded and priced, so the operator sees what each conversation costs. Public agents have daily spending caps and visitor limits.
It runs where it needs to. In the cloud for phone and web agents, and on device for private personal agents.
The operator's view
Each agent comes with dashboards:
- Leads: the people an agent spoke with and the commitments waiting for a person.
- Calls and chats: every conversation, with what the person needed.
- Interest: which parts of the documents people actually ask about, and what they asked that the documents could not answer.
- Economics: what the agent costs to run, by conversation and by model.
- Config: the agent's settings, its phone and chat switches, and how it reaches the operator.
Who AgentOS is for
Organizations with a body of knowledge that people need to get at: a company's services, an agency's rules, an archive, a catalog, a body of research. Blacksky's core offer is turning a large collection of material into something people can talk to, on the phone or on the web.
How an engagement works
Blacksky builds the agent: curating the documents, writing the persona, setting what it can do, and connecting the phone number or the website. The organization owns what is built. Hosting runs on Black Cloud as its own itemized line, and memory on Prism. See How an engagement with Blacksky works for pricing in shape and typical timelines.
Common questions
Can we bring our own documents? Yes. That is the starting point of every agent.
Can the agent connect to our systems? Yes, through tools set up for the agent, scoped to what it needs.
Do you train models on our documents? No. Blacksky uses documents for retrieval, so the agent can find and cite them, not to train a model.
Can we take it with us? Yes. The configuration, documents and data are exportable.