A base rate for every agent you run, and published rates for the sessions they run and the memory they keep. Bring your own model key, so inference stays at your provider's own rates and never passes through us.
An agent is one worker your fleet can run at the same time. Ten agents means ten tasks in flight at once. You set the number, change it whenever you like, and are billed to the day.
What that rate buys is shared memory, per agent that uses it. One project memory is written once and read by your whole fleet, so every agent starts knowing what the others already worked out — instead of rediscovering it, in tokens, every session.
On top of it, meters bill what your agents actually use — the sessions they run and the memory they keep — at the published rates below, from the first unit. A fleet that ingests and processes large amounts of data is scoped with us first, so the bill matches the work rather than arriving as a surprise.
Models are not on this bill. You bring your own key, so inference is billed to you by your provider at their rates. We never resell it and never mark it up. Run the model on your own hardware and there is nothing to bill.
What is metered is in units you already count: the tasks your agents take on and the memory they keep. Every published rate is here, and every one bills from the first unit.
| Line | What it charges for | Rate |
|---|---|---|
| Agents | Shared memory, for one agent that reads itYour project memory is written once and read by every agent. Each project is isolated, with its own memory; run as many as you like. | $10.00 per agent, per month |
| Sessions | Each task an agent takes onOne session is one piece of work, start to finish. The nightly review that keeps your memory true is not a session and is not charged for. | $0.04 per 1,000 sessions |
| Memory | What your agents have remembered, kept for youMemory accumulates and every version is kept, so this is the line that grows with you. | $1.00 per GB, per month |
| Networking | Reading your memory back outYour memory lives in storage that charges nothing to serve it, so this line is zero and stays zero while that is true. | $0.00 per GB |
| Processing | Agents that ingest and work through large bodies of dataA fleet working on a codebase is one thing. A fleet ingesting a warehouse, or holding a standing role over one, does substantial work between sessions. That is scoped and priced with you before it starts. | Priced with you |
Because no two fleets use the same amount. An agent that runs ten times the sessions bills ten times the sessions, and a project holding ten times the memory bills ten times the memory. Folding that into an allowance would mean charging the ordinary customer for the heavy one and calling it “included”. You pay for what you use, and the rate that turns your usage into a number is printed above.
More than a fleet working on a codebase — and we would rather tell you the number than surprise you with it. Agents taking on a line-of-business workflow, or holding a standing role over a body of data, do substantial work between sessions: ingesting it, making sense of it, and going back over it as it changes. No honest per-unit rate covers that in advance, so we do not publish one. Tell us what the workload looks like and we scope it with you before you commit to anything.
Yes. The fleet is reachable from the command line and from your own scripts, through the clients we ship, so an agent you drive yourself shares the same memory the rest of the fleet is writing. You are not restricted to one console.
Any of them. Anthropic, OpenAI, Google Gemini, Amazon Bedrock and Hugging Face all work directly. So does Ollama, and so does any OpenAI-compatible server — vLLM, LM Studio, LiteLLM and the rest — which means you can run models on your own hardware and Dream Maker reaches them the same way.
Because it keeps your inference at your provider's own rates, with every token visible in your own dashboard, billed to you directly and never marked up by us. It also means every line on the bill above is something we actually run for you, with no inference cost averaged in underneath.
A project is one codebase with its own memory, and it runs in its own space. A team working on your API sees the memory of that project alone unless you connect it to another. Projects are not charged for; agents are.
You are charged for the days they run. Adding capacity for a busy week costs a week, and removing it stops the charge the same day. Your memory stays exactly as it was either way.
No minimum and no term. One agent is $10.00 plus its meters. Stop running agents and both the base rate and the session meter stop; memory keeps billing at $1.00 per GB per month for as long as you want it kept, and nothing at all once you delete it.
Yes. Dream Maker runs inside your own cloud with your own keys, and we never see your code or your memory. Self-hosted deployments add SSO, audit export and custom governance policies. Talk to us about your requirements.
Dream Maker reads your project and proposes the goals it finds there. You approve the direction, and your fleet is working within the hour.